clawtributors

[lunar granite]

Soooo should we share PRs here? 🙂

Would love to contribute more

#general is so hectic right now hhh

[clear plank]

i honestly don't know yet 😄 the idea is to somehow focus discussions on PRs and improve contribution workflows yet, whilst still keeping them lightweight, and reducing Peter being a bottleneck. let's just try and share PRs in here, keep discussions in threads, and iterate on the approach and see how it works out 🙂

[celest merlin]

Yes or also chat about what you wanna add.

[clear plank]

Probably the meta-discussion is more important than the actual PRs: how can we figure out how to have an effective flow of focused and quality PRs, at the speed of AI, and outsource as much as possible to Clawd to review/triage etc for us.

[tall prairie]

when 10 people becomes 10x it's 100x the code

[lunar granite]

So I have a discussion about the Mac app in remote gateway mode overriding my gateway configuration and breaking my setup - opened a pr for that but was wondering if it’s intentional

Maybe it’s more of architecture/technical question

[clear plank]

Current challenges: we have lots of PRs, move very fast, have lots of open issues, its hard for everyone to keep the context in their heads, and keep the project stable.
Also, we need to keep PR quality high, focused, and non-slopified, and figure out user intent & prompting history (so people can re-implement PRs if needed with additional context). Since in the age of AI, PRs are somewhat disposable - its sort of more idea requests with context attached.

[tall prairie]

honestly i feel like if the code can evolve so that the core of the project becomes it's own thing that Peter works on while making everything else extensions that can be installed and developed/managed indepently is a good direction. Think about wordpress, it's core is separate but extensions ahve their own lifecycle and management and are "installed" into wordpress.

then farm out the less important parts to individual contributors that can own their piece like clawdhub and can refine it like making mobile look good and all that jazz

clawdhub becomes the non-core hub, finding isntalling extensions/skills etc

[clear plank]

yeah that's a good direction IMO.

[tall prairie]

even find some vibe marketers that can do things like maintain the clawdbot.com page, docs site etc

i'd also get a discord mod team to manage the influx of people coming in

[jolly wolf]

That’s us! :)

I used to run servers professionally so I’ve been helping out here with it

[clear plank]

thanks for the tips @tall prairie ! i like your ideas. i'm also exploring a "clawd as maintainer" idea. which will hopefully unlock a few things. will experiment internally before letting it out in the wild

[tall prairie]

probably add to that, maybe add a "classification step to a pr" as in, is it high risk or low risk, high risk goes to Peter, low risk goes to maintainers

but yeah the extension framework to push off those things to non-core is good then you aren't just telling people no, you're saying go build an extension!

[clear plank]

exactly 🙂 empowerment vs gatekeeping

[jolly wolf]

honestly ive been working on the connectors in such a way that they're easy extensions, at least discord and slack are

[flat lark]

any feedback on this Plan to add native Google Cloud Vertex AI auth for Anthropic model use? the existing vertex-ai integration appears to only support Gemini models: https://gist.github.com/mikem-coalesce/9ff71d83451f8467c5b4bc55f2d82e9d

[woeful portal]

whats the prefered size of a pr. It's a bit difficult to chop down bigger changes to make them reviewable?

[wicked ore]

this repo is not for the weak of mind

i just can't deal with contributing

[jolly wolf]

ive gotten extremely good at rebasing

[hidden fox]

I've written this because I've had this branch stack of 12 branches with 13k lines at work. It dry-runs the rebase in a plan.yml, and only after conflicts are resolved, does it the rebase in a worktree. Also resets all branch tips. Don't use in production just yet 😉 but maybe useful for gnarly updates. Currently hard-coded to feature/* branches but that would be fixable for sure (PRs welcome :D)

https://github.com/manmal/git-restack

[steel acorn]

Submitted a PR for supporting topics + groups.
Looked at existing PRs ... a bunch lol
Then looked at issues... even more haha

No rush on my PR, its working on my end lol just figured I'd pull latest and run that but will wait till my changes get merged. Still having fun with it

[raven prairie]

It's all good! We're just getting started and just needs to be some organization, issues and prs with tags and labels, start proper version control for stable release, start nightly builds and bleeding edge repos, organize skills and the skills hub a bit more with perhaps even a peer review, delegate some mods. Lots of volunteers is awesome.....but without direction is chaos. 🙂 Fun...but chaos.

the energy is 🔥 and I'm loving the Clawdverse

we should also setup a github bot to do basic pr checks etc...if it's not already set?

[jolly wolf]

WIP right now actually :)

[raven prairie]

Plus with a proper roadmap (or just the bones of one) it's easier to triage PR's and also delegate to even clawd to merge if passes

[clear plank]

working on it 🫡 its gonna be br00tal

[raven prairie]

See I'm very glad to know that, because it saves me from duplicate effort. I wasn't aware and I'm glad to hear it!

[clear plank]

[raven prairie]

That's just the thing as above with 3 PR's and counting now.

We organize our clawds to all work on different things and we're gonna really start moving.

[clear plank]

yeah absolutely. im actually a bit scared of the swarm of clawds idea but yeah. it will make things less chaotic here and easier for us to maintain everything

[tall prairie]

The holidays are over. I hope things will slow down a little bit so you can catch your breath.

[clear plank]

i think peter is the most affected here ❤️ but we are working hard to unbottleneck him and enable the org to scale (and reduce the chaos on his plate 🙂 )

[celest merlin]

What a day yeah

[steel acorn]

Was looking into that auth refresh issue a couple people posted. I believe it was this commit https://github.com/clawdbot/clawdbot/commit/2dd6b3aeb24a6d6176bee58aad7e0e365b3c11bf#diff-583a38373c4263b97b76183bef1664f5981f2d5aff9717924fa3e9fd777c8e15

Going to submit a PR to show a possible fix. I still need to debug it though.
If I run across the issue (since I just updated) will run my the feature branch as my gateway to see if that resolves it.

Maybe not an issue but would suck if everyone has to keep refreshing their auth process.

[untold pollen]

Oops I completely missed this channel. Made a post here instead: https://discord.com/channels/1456350064065904867/1458322349278625902

[steel acorn]

Submitted PR. The change there had it hardcoded to :default but that writeOAuthCredentials method could save it under anthropic:email.com

Just a mismatch and I guess of when someone initially set up their account without an email? idk

[untold pollen]

oof, 2 people made the same Telegram concurrency PR

[steel acorn]

lol everyone is fast

[marble notch]

Hey everybody 🙂 Just joined and would be glad to participate in development.

I'm particularly fond of telegram, so I'm curious to see how we have streaming there.

From user's perspective I see that it:

a) doesn't stream
b) sends many messages instead of single one.

Any insights and pointers for me?

I see this PR: https://github.com/clawdbot/clawdbot/pull/236/changes, but it's not that easy to figure out.

As far as i understand, we actually do "streaming" but chunking the output and sending separate messages? Anyone working on proper implementation — meaning sendMessageDraft streaming and forumTopic mode (support for topics in direct messages, don't mix it up with group topics)?

@celest merlin heh, saw your reply in X, that you're "on it" for streaming :)) So, you're taking lead on that?

Because that would be way cooler and robust experience that chunking and sending lots of separate messages (which also triggers lots of notifications on my devices)

[steel acorn]

Nice. I did see that in logs but I thought it was just seeing the text just
Getting outputted as it was being wrote out

[celest merlin]

just vibing streaming

[celest merlin]

Forum? 4!!

[untold pollen]

@alpine portal we need to keep track of who's doing what haha

[clear kraken]

heyo

[harsh panther]

Morning 🙂 I had trouble the mac app discovering my gateway so made some fixes here: https://github.com/clawdbot/clawdbot/pull/375

[outer birch]

Just a small change but I had codex add the day of the week to the system prompt (not only time and date) because my clawd would always mess up the day of the week (told me it would be tuesday today in my morning brief) https://github.com/clawdbot/clawdbot/pull/378

[clear plank]

merged, thank you @outer birch
codex review approves

[tall cairn]

my first pr. I am so glad I found this project 🥳

https://github.com/clawdbot/clawdbot/pull/380

[marble notch]

fml, i just managed to go and begin working on strreaming in telegram, ran git pull and it's already done 🙂

this is singularity. obviously, in 3 days clawd on dozens or hundreds computers would start to self-improve and we're done.

[grand aurora]

Nice, I have been working on this. I opened a PR just now: https://github.com/clawdbot/clawdbot/pull/381

Took me quite a while burning the midnight oil...

It seems to really help with token burn. Should we discuss here or in a GH discussion about this approach?

I followed what opencode, codex, and claude code all seem to do. I didn't find a blog post about codex doing it, but if you reads its thinking monologue, you see it complain that it loses some tool output, so they definitely do it.

How my pr works: once enabled, tool calls which are first truncated (well, keep the first and last parts of the tool call, this is what twitter says codex does aggressively) and after that pass, goes and prunes oldest tool call results (not just truncate them).

Need to update your local config and un agents set:

"contextPruning": {
      "enabled": true
    }

I also tested it with super aggressive settings:

    "contextPruning": {
      "enabled": true,
      "softTrimRatio": 0,
      "hardClearRatio": 0,
      "minPrunableToolChars": 0,
    },

Whats missing is some more testing longer term to see if I notice clawdbot being any dummer, and how aggressive to tune it. I think compared to most llm cli's we can be pretty aggressive since probably we're all just having short convesations with clawd that doesn't often need info far back in the history.

Hopefully this 'fixes' token burn and lets us all enjoy more Opus time with clawd 🙂

[clear plank]

awesome, thank you so much for diving into this

[untold pollen]

Building it as a pi extension is perfect

I’ve been testing my own version locally so far and it works really well

[grand aurora]

its 100% AI slop but a lot of back and forth to try and massage it into something nice. I had a really shitty version a few days ago to verify it works, so tried to make it something actually mergable here

But its codex slop so 🍷 👌

[clear plank]

i dont have capacity to look right now (i'm trying to work on the meta contribution workflow so we can unblock everyone instead of firefighting) but this kind of contribution is exactly what we need, thanks max

[grand aurora]

From my reading of the tweets (thanks for the mention @clear plank ) and what else I find, it seems all the other harnesses do this. That would explain why it seems pi (and thus clawdbot) uses "a lot more token" than we're used to.

[lunar granite]

this is awesome! trying it now as well

[raven prairie]

I noticed someone changed tsx for bun in the build yesterday.

[lunar granite]

🤞 @grand aurora

[raven prairie]

this scares me. fear the sycophancy.

[clear plank]

one thought i just had- if you didn't do it already, make sure to ask codex to do some brutal self reviews on big code changes, and ask it if the feature is appropriately architected and sliced and diced, with everything living in the right place. (this is basically what other people will do on the PR). ask it about potential footguns, or if it would implement the PR differently with the knowledge it has now. that will hopefully make your changes as bulletproof and likely to be upstreamed as possible. (this is also a general tip for anyone contributing. apologies if it's old hat or obvious)

[raven prairie]

https://github.com/clawdbot/clawdbot/issues/384
can we get a clawdbot setup so that we can @ them in comments and have labels added (and types) we need to be able to triage and escalate/de-escalate

[clear plank]

we are working on this 🫡

[raven prairie]

who is "we"

I'd like to help and it become "us"

[clear plank]

actually more advanced 😈

[clear plank]

me 🙂

the idea is that we just spin up a clawdbot instance with access to github + discord + also has codex so can code things for us, and is basically "clawdbot as maintainer"

codex is currently brring on it as we speak, the architecture is there, i just need my builds to succeed

the clawdbot will have memory of the project, architecture, priorities, project philosophy, etc, and can help guide us basically

[raven prairie]

Ping me when it's done, I'll test it

Also, who has mod access on gh for labels (human-wise)

[clear plank]

thanks, will do. if there is parallelizable work i will do so; what are your areas of expertise?

[raven prairie]

i'm fairly deep technically, and done a fair bit of volunteer work, helped build some cool things, lots of stuff i guess, but can handle any level of tech

lived a lifetime in open source

[smoky ether]

@raven prairie are you the Mautic David Hurley?

[raven prairie]

found out

[smoky ether]

lol! we met once in person for like 15 seconds at Acquia office, very close to your last few days 😄

[raven prairie]

I absolutely remember you! We had a whirlwind couple of months. I remember we had a great trip downtown also for our offsite dinner.

[smoky ether]

yes!!! we did. Well, don't hold all that and Acquia against me (still there). Glad our paths have circled back!

[celest merlin]

looking. great dig!!

[raven prairie]

Hahah, good to know 😉 Yeah, i have several really good friends still there.

[grand aurora]

One bug: need to also exclude pruning the messages before the first user prompt, top prevent pruning the tool calls to read SOUL.md, etc...

[grand aurora]

One bug: need to also exclude pruning

[jolly wolf]

We’ve had a lot of recent clawdributors lately! If you’re missing the role, ping me and we’ll get you sorted!

its our way of crediting and listing people here as well as on the gh repo!

[junior spade]

Hi, I recently got this pr merged: https://github.com/clawdbot/clawdbot/pull/409

[lunar granite]

question - wdyt about adding a feature to run skills on nodes that connect to gateways in remote-gateway mode.
My setup is - gateway runs on rpi - rather limited but got a lot of things configured (gog, many skills.. etc') - and a macbook running clawdapp connected to the gateway in remote mode - to my understanding (and did several rounds of code-reading and codex investigation) - the gateway can't route skills to nodes, so if I want to use apple-reminders/similar mac-only skills - I can't right now (clawdbot finds clever ways around it - using system.run exeucting remindctl for example)

I was thinking about doing proper skills broadcast from nodes and routing to them from the gateway when needed - if I'm missing something please tell me

(my mac is my main working horse but it's sometimes offline - so I want the rpi gateway to use it for the cool things clawd can do)

and by adding I mean - I'll work on it ofc heh

[faint harbor]

I was hoping for the same thing. After 15 years of distributed computing programming tho I'm more in love with centralized because it doesn't come with a whole class of edge cases. Maybe with AI able to adjust, the classic "rpc endpoint went away, now what part of CAP do I choose?" issues aren't as much of a problem.

I'd think it would be quite a common use case.

1. User has laptop they work on and close regularly.
2. User has cloud server to host their gateway (probably linux).
3. User should still be able to chat with gateway, when laptop is closed.
4. User should be able to expose macOS functionality as RPC to gateway.
5. User should be able to queue work for the laptop for when it connects back to the gateway again.

[dusky blaze]

Pinging You 🙂

[jolly wolf]

send one of your pr links so i can note that i have you!

[dusky blaze]

https://github.com/clawdbot/clawdbot/pull/377#event-21895950332
https://github.com/clawdbot/clawdbot/pull/218#event-21858634819

and thank you for your work on all of this, you and @celest merlin really are making this a magical experience 🙂

[frank berry]

Where does it show? I love digital street cred

[jolly wolf]

Gradient pink name here and listed in the member list

[frank berry]

Oooh ahh. Okay, load me up plz. PRs 126, 127, 190 and 193

https://tenor.com/view/lobster-yes-happy-excited-cant-wait-gif-14356908674053187701

[tardy sparrow]

I think, i've got about 7 pr's merged?

1. #88 — feat: Add Google Antigravity Authentication Support
   https://github.com/clawdbot/clawdbot/pull/88

2. #124 — fix(telegram): add textLimit to block reply chunking
   https://github.com/clawdbot/clawdbot/pull/124

3. #167 — fix: Antigravity API compatibility and Gemini thinking tag leakage
   https://github.com/clawdbot/clawdbot/pull/167

4. #269 — feat: Multi-account OAuth with round-robin rotation
   https://github.com/clawdbot/clawdbot/pull/269

5. #280 — fix(cron-tool): use generic object schema for job/patch
   https://github.com/clawdbot/clawdbot/pull/280

6. #281 — fix(auth): prioritize round-robin over lastGood
   https://github.com/clawdbot/clawdbot/pull/281

7. #312 — fix(build): import tool-display.json instead of fs.readFileSync
   https://github.com/clawdbot/clawdbot/pull/312

[jolly wolf]

yup already got you!

i got most of the people who had their discord and gh usernames the same as of last night

[tardy sparrow]

thanks 💌

[tall cairn]

@jolly wolf could you also have a look if it was worth to be mentioned?

[jolly wolf]

Peter hath deemed it so!

but yeah any contributions will count :)

[faint harbor]

I don't know if it's just me, but I'm working on fixing some of the issues with the Voice Wake on the macOS app. Specifically around the Start test/stop, etc.

[wooden parcel]

First PR 🎉

https://github.com/clawdbot/clawdbot/pull/430

Second PR with fallback error with roles for GLM specific model

[versed kindle]

serious question about teams implementation

I've seen how well this thing works on discord, and I want to have the same level of UX

i.e. you can ask "what did we speak about 1 year ago, summarize", and it can reach entire history to do that
OR even more basic, it can read messages sent when the bot was offline, so people don't have to repeat themselves

unfortunately, teams doesn't have roles like discord. to be able to read any message, you have to give extremely wide permissions:

⏺ For historical message retrieval via Graph API:

  | Permission              | Type        | What it does                               |
  |-------------------------|-------------|--------------------------------------------|
  | ChannelMessage.Read.All | Application | Read all channel messages (tenant-wide)    |
  | Chat.Read.All           | Application | Read all chats including DMs (tenant-wide) |

  Both require admin consent and grant access to the entire tenant (not granular).

once this is granted, and the bot has access, it can read in all channel messages or dms in the whole tenant

some companies might opt in, some would probably not. for those who opt in, there are enormous advantages

given that company trusts the infra, the best solution looks something like this:

⏺ Option B: Graph API Proxy Gateway

  ┌─────────────────────────────────────────────────────────────┐
  │                        Your Tenant                          │
  │                                                             │
  │  ┌─────────────┐     ┌──────────────┐     ┌─────────────┐  │
  │  │   Clawdbot  │────▶│  Graph Proxy │────▶│  Graph API  │  │
  │  │  (no token) │     │  (has token) │     │  (tenant)   │  │
  │  └─────────────┘     └──────────────┘     └─────────────┘  │
  │         │                   │                              │
  │         │                   ▼                              │
  │         │            ┌─────────────┐                       │
  │         │            │  Allowlist  │                       │
  │         │            │  Config     │                       │
  │         │            └─────────────┘                       │
  │         │                                                  │
  │         ▼                                                  │
  │  ┌─────────────┐                                           │
  │  │   Teams     │  (real-time via RSC webhook)              │
  │  └─────────────┘                                           │
  └─────────────────────────────────────────────────────────────┘

  How it works:

  1. Graph Proxy is a small service (could be a Cloud Function, separate process, or even an MCP server)
  2. It holds the ChannelMessage.Read.All token
  3. Clawdbot calls it with: GET /messages?team=X&channel=Y&since=timestamp
  4. Proxy checks config: "Is clawdbot allowed to read channel Y?"
  5. If yes → fetch from Graph API, return messages
  6. If no → return 403 Forbidden

so locally I already have basic functionality working, including clawdbot side config such as requireMention, which channels to receive, etc.

wanted to ask before I move onto something more complicated like this

Would this look good in the context of this project? Or is it too out of its place?

I can build the graph api gateway separately, and use it in the project. or it could be a component directly under clawdbot

wdyt @celest merlin @jolly wolf? should I just implement away?

[tall prairie]

Graph api is so bad

I deal with it at work 🤮

The whole tenant thing is a major pain point

[versed kindle]

it's surprisingly pleasant compared to having to do actual clickops

[versed kindle]

it's communism in digital form 😄

[tall prairie]

Haha

[versed kindle]

engineered inefficiency

when you have the api at least documented, you can at least automate stuff

[tall prairie]

The whole design of it is annoying. You either have access to your own stuff or the entire tenant and there’s no middle ground.

Not sure what they were smoking when they designed it

[versed kindle]

I was trying to upload my teams app to my tenant for 1 hour, and I was getting "Something went wrong" in the UI

you don't even get an error message. and teams app doesn't give you network responses

found out what's wrong by finding some accursed admin panel in ms teams website and using the inspection window

it was the fucking icon

this is arc agi-4 IMO

when we have cheaper and better multimodality, I will just give the model computer access and let them deal with it

not to mention that the "app" is a fucking json + images bundle you can download as zip, which has to go through some "app validation" for some reason

and app validation seemed to be manual human based, because it can't take that long to inspect a json file

anyway too much yapping but I wanted to record somewhere how many steps you have to go through just to create a teams app. will need to create a guide or automated client for it some time

[steel acorn]

Nice to see movement! Got 2 PRs in for now. Will try and help tackle issues every night

[celest merlin]

Thanks all ❤️

the human merge button is very thankful

[kind crater]

hi i just made a patch to make the browser tool work on windows (im using bun)

do i create an issue first and then open a PR on that?

[raven prairie]

Yes that would be great!

Issue + linked PR

[faint harbor]

I think I've got the Voice Wake feature a lot more stable now: https://github.com/clawdbot/clawdbot/pull/438

[kind crater]

https://github.com/clawdbot/clawdbot/pull/439

[celest merlin]

folks if anyone did a PR that I used/merged but where codex messed up, hit me up so I can fix it. ❤️

[faint harbor]

Pretty sure I'm not in there

[heavy rune]

Excited to have my first two PRs merged in today, #364 (https://github.com/clawdbot/clawdbot/pull/364) and #433 (https://github.com/clawdbot/clawdbot/pull/433) , looking forward to more to come, and super excited about the pace at which this project is moving. You're all doing an amazing job. 🙂

[raven prairie]

How's it looking? Any luck?

[celest merlin]

Folks - please hand review and hand-write code when you submit to pi-mono and REALLY spend time on it. Marion has quite a different approach and he just sent me a super pissed message.

[smoky ether]

was myfault or soemthing else? (btw didn't actually intend for my bot to submit a pi-mono PR, it just decided to)

[jolly wolf]

Honestly this applies to most repos on the internet, clawdbot is the oddball one here

[smoky ether]

actually I think my bot only opened an issue on pi-mono, not a PR 😓

nvm I see

honestly, and I know this sounds weird, you might just want to fork his repo and have clawdbot build off the fork. setup a daily sync or something to mirror.

Everyone's clawdbot and codex is going to walk the dependency tree and recommend filing PRs there if it thinks there is a bug. Will probably keep happening more and more as clawdbot gets more installs.

[celest merlin]

I'm not gonna maintain a fork.

We already have ugly patches, and it has to be, I'll work and clean up and upstream

[smoky ether]

Was thinking more of a mirror then a fork, just to act as a shield to stop bots from self navigating to pi mono repo.

[celest merlin]

Please don't even try to upstream that

He's already pissed at the quality.

I'll do a patch for now... ugly a f but I see so many folks complaining so what can you do 🙂

codex is alrady weaving it in

[frank berry]

Thanks Peter 👍

[frank berry]

I enjoyed reading his blog the other day about building Pi. Picking a name that wouldn't turn up well in Google, only to be put front and center in front of the worlds maddest vibe coders by a space lobster 😂

[silver crown]

Hi Everyone, I am Samarth. This is a very exciting project. happy to have my first PR #387 merged to the clawdbot project today. Looking forward to clawdribute more 🙂

[celest merlin]

Welcome, clawdributor!

Oh we hacked together... we just have different philosophies.

I'm more builder he's more thinkerer

[reef atlas]

love "thinkerer"

reading your messages have got me scared to even try to submit a PR to him lol

[wooden parcel]

Happy to have this merge ( https://github.com/clawdbot/clawdbot/pull/430 )

Noted for the pi-mono repo, I had a weird issue with GLM and user role error ( I’m on the coding plan) trying to replicate the behavior but will refrain opening anything there without being able to replicate before 🙂

[celest merlin]

thank you!

[silver crown]

Hey Peter. Tackled two issues -213, 417 as the mac menubar app kept crashing. https://github.com/clawdbot/clawdbot/pull/473

Also the CI keeps failing - lockfile is missing markdown-it dependency.

[silver crown]

added a PR to update a few interesting use cases i saw on #showcase-old - https://github.com/clawdbot/clawdbot/pull/476

[topaz fable]

added PR https://github.com/clawdbot/clawdbot/pull/475

[tall cairn]

https://github.com/clawdbot/clawdbot/pull/480

shot my config twice now during testing and dev. so i made a backupscript 😉

[raven prairie]

Necessity the mother of invention. 😉

[versed kindle]

#general message

whoops wrong reply

teams PR is ready for review

[vocal merlin]

Oh Mario

[steel plover]

Hey all! I followed the contribution guide and posted a PR. After the PR is live, do we need to engage with reviewers for awarness? Curious about what folks do here after publishing PRs

[clear plank]

currently its chaos and peter-driven prioritization. we are hard at work on getting our own clawdbots to make it a more structured process.

[raven prairie]

PRs get merged without gaining approvals at this early stage, so no need to get testers, but....we ALL like knowing about new features and fixes! 🙂

[raven prairie]

Along those lines have we gotten that bot first pass finished on the GitHub repo assistant? Surely clawdbot is done churning on it by now 😉

[clear plank]

long story but uh, give jeff bezos money instead of the hetzner guys. codex works better on it.

@stuck spade is currently in its bootstrapping phase. making sure its instructions are correct. and the hivemind is online.

it will work today

its going to be br00tal

@stuck spade is soundtracked by https://open.spotify.com/album/6UrvQgyblrOuvJytote1vu?si=lJJmU7KsSiKCattvUhKzjg

[raven prairie]

Sounds good! Where's its code going to live?

[clear plank]

for now privately since this is basically clawdbot as a service (for maintainers), which peter expressly does NOT want people to make. but really it's just the nix-clawdbot package thrown on AWS. but this could change rapidly (if peter says its fine to open it up). im being cautious for now

[raven prairie]

...

noted. I'll begin working on an open source version that we can commit to the Clawdbot repo!

[clear plank]

(its private in the clawdbot org btw)

[raven prairie]

Sure! I'll make an open source one that follows mit license and is available to all

[raven prairie]

I don't have access to that org, I don't think many others do either

[clear plank]

this probably falls under the peter saying "please do not do this part" tbh. so i would check with him to make sure everyone is aligned.
its a bit difficult, because things that make it easier for normal people to run clawdbot (good) make it easier for 3rd parties to commercialize (not desired). (so this is why i'm keeping the tool internal for project maintainers until peter weighs in on what he wants)

[raven prairie]

I hadn't heard that from Peter about not wanting to provide the services we create with Clawdbot from the org.

But it's MIT license so if it's something that he doesn't want to see from the organization that's completely fine, it will be hosted by private party orgs on GitHub

I completely understand if he doesn't want the project to feel commercialized by having services provided or offered

[clear plank]

yeah i'm paraphrasing here from what i've heard him saying. its his decision not mine 🙂

[raven prairie]

Oh, it's totally fine. The product is going to be commercialized. It's just whether or not we share from the org how we do it in our version or thinking.

I reached out to Peter already a bit ago. And we've chatted a bit, but he is understandably overwhelmed right now and was unable to meet. I'll keep working away!

How does the maintainer team get selected? What's the process to join?

[steel plover]

Netflix style 😂

[raven prairie]

It works! As long as the feedback and fix loop is tight

[steel plover]

For awarness, Control UI bug fixing: https://github.com/clawdbot/clawdbot/pull/445

[raven prairie]

Awesome!

[steel plover]

Oh nvm, it requires maintainer approval! It can wait

[raven prairie]

Yeah Peter has to actually push the button

[hasty forge]

did you guys fork the repo to make prs or just creating branches on the main repo?

[raven prairie]

I create my own fork of the repo, branch for each fix, and then submit prs to main repo, peter merges direct to main, then I delete my branch after merge

[hasty forge]

thanks

[smoky ether]

curious to know more later on AWS vs hetzner (never used it) but a VM is a VM right?

[clear plank]

yeah its just codex knows how to write better terraform/opentofu code for AWS

since more people use it

[smoky ether]

ohhh - not about how it runs on the VM, about how the infra is managed. That makes sense

[clear plank]

yeah burned a day while the codex did dumb shit trying to figure out how to get VMs working isntead of falling back to boring build AMI -> push AMI -> run pipeline lol. whoops.

[raven prairie]

I've been there.

[raven prairie]

Is #clawhub-skills webhook endpoint ready to hook into Clawdhub? Would be great for it to notify on new skill posts @clear plank or @jolly wolf likely best guess?

[tall prairie]

Krill isn't in this channel, usually in #old-help, he specializes in searching and answering questions about the codebase.

[raven prairie]

Who's looking for that crustacean clanker?

[split nexus]

Added PRs to help with onboarding/discovery:

• /commands: https://github.com/clawdbot/clawdbot/pull/497
• /usage (alias for /status): https://github.com/clawdbot/clawdbot/pull/492

[bronze arrow]

Added this PR as I ran into getting clawd to send messages back in slack threads. No conflicts at the moment

https://github.com/clawdbot/clawdbot/pull/464

[vernal tundra]

Added this PR cause Clawd Whatsapp has a hard limit in the code of 5Mb for a PDF, at least let the user configure a higher limit, I like Clawd to read long documents and create nice summaries for me https://github.com/clawdbot/clawdbot/pull/505

[smoky ether]

Found an issue when discord is slow, bot will go offline forever https://github.com/clawdbot/clawdbot/issues/504

and it's partner for messages failing to send https://github.com/clawdbot/clawdbot/issues/507 @clear plank not sure if this is overlapping or not with https://github.com/clawdbot/clawdbot/pull/503 but I don't think so

[lunar granite]

Continuing the thread ^

I started a discussion around implementing "Node-hosted Skills" - it's a rough spec I did with codex, it suggested two phases for implementation.
High level goals -

• Allow macOS-only skills to run via the macOS node while the gateway stays on Pi.
• Preserve current behavior for gateway-local skills.
• Explicit, auditable routing with clear fallback behavior.
• Keep security model consistent with node pairing + permissions.
  https://github.com/clawdbot/clawdbot/discussions/508

If it sounds interesting I can start working on the implementation
(my current setup - rpi as main gateway and mac as node, is nice but lacking since I can't use a lot of mac skills - while clawdbot is smart enough to go around that - mostly by using node.invoke + system.run - would be awesome to get "native" skill execution)

[smoky ether]

could SSH just be used here? I don't have clawdapp etc b/c all linux boxes, but I do have my bot jump between boxes and run things like crazy

[lunar granite]

It does that right now (using node.invoke + system.run - basically executing stuff inside the node) - but it kinda defeats the purpose of having a specific skill written (like apple-reminders or other darwin specific things) - clawd is smart enough to do a lot of things but I'm wondering if I'm missing something in my setup because skills around bound to my rpi - lame linux ubuntu container

and we already have node specific things that are implemented in the mac app (screen capture.. etc')

@smoky ether adding to what you wrote - SSH alone will def work but imagine having each linux machine exposing the actual skills it has (one is connected to a webcam, another got your google credentials)

[clear plank]

ok @smoky ether 503 is merging. whats next 507 or 504?

[tall prairie]

Guys, security is kind of a moot point when the agent can write and execute arbitrary shell commands

Just saying

Don’t give the agent access to stuff you can’t risk

[lunar granite]

Right now clawdbot running outside of a container can basically do whatever it wants on the system afaik no?

[tall prairie]

It will do whatever instructed to do

[lunar granite]

yeah ofc given instructions

but I've read horror stories of agents doing mistakes when removing files like "rm -rf folder1/* ~/"

[smoky ether]

either - they are seperate issues

[smoky ether]

hence mine having its own google and github account

[faint harbor]

@jolly wolf can I get Clawtributor role? My PR: https://github.com/clawdbot/clawdbot/pull/438 Thanks!

[clear plank]

codex brrs. time for dinner

[faint harbor]

We need Clawd to be able to restart Krill when it starts repeating itself

[onyx sparrow]

is there a particular reason anthropic auth doesn't check/request keychain as a fallback on mac?

have had to re-auth 1-2x a day

[lunar granite]

could be the refresh token collission issue - this was merged yesterday https://github.com/clawdbot/clawdbot/pull/374

[onyx sparrow]

CC doesn't use .credentials.json on Mac, uses keychain

[daring saddle]

I had same issue and it was collision with the refresh tokens for claude code and clawdbot.
I got it working by generating a long lived access token with claude setup-token.
Then added it to config like this:

      "type": "api_key",
      "provider": "anthropic",
      "key": "sk-ant-xxx"
    },```

[brittle mulch]

Anyone else having refresh issues in the TUI?

[mighty tapir]

The #1457462547329454312 needs to have links to the website, docs & clawdhub.

[jolly wolf]

It needs an update overall tbh

[tall cairn]

Anybody help me with a feature!

Having sandboxed agents (family, work) and main agent. I want my family to use some custom tools like gurkerlcli for grocery shopping. cannot get it done. Any ideas, tried to slope a custom-tools implementation where allowed tools will run on host but triggered by sandbox.

Any suggestions?

[raven prairie]

I was just going through the docs and wondering if it was time to do a refresh? Because I believe they've gotten out of date.

I was also wondering if we should set up a channel here in Discord specific for #docs

[jolly wolf]

Please! https://discord.com/channels/1456350064065904867/1458532902999556188

[jolly wolf]

Here is fine, I’m trying to avoid channel overload

[faint harbor]

I think someone (maybe @clear plank or @jolly wolf ) dropped a prompt they use to review a PR before merging but I can't find it. Maybe something about refactoring knowing what we know now? Does anyone have that prompt so I can run it before submitting PRs?

[jolly wolf]

Just /review in codex

[faint harbor]

Ok

[jolly wolf]

I do more both with codex and manually, but that’s the first step because it usually catches most things

[raven prairie]

Can you also answer my question about the skills announcement channel and whether or not it got finalized and hooked up to new listings at Clawdhub

[jolly wolf]

Missed that message, it’s hooked up and supposed to be going but broken

PR welcome, ping me if you have one for that

It also doesn’t seem to be pushing to gh on publish but the cron is fixing that

[raven prairie]

I don't know where the code live so I can address it, or rather what's the web hook url the listing should be posting to

[jolly wolf]

It’s an env variable

Webhook urls should be treated as secrets

[raven prairie]

Right, so I'm not sure how I can test things and fix the code in the hub

[jolly wolf]

go into a server you own (i use my clawdbots server), make a channel, go to the settings for that channel, there's an integrations tab, you can make a webhook there

[raven prairie]

Noted

[lunar granite]

https://github.com/clawdbot/clawdbot/pull/555
fyi small fix gateway doesn't compile now (due to https://github.com/clawdbot/clawdbot/commit/db222070)

fixing linter stuff

okay saw it was fixed already

[celest merlin]

Pls don't send fix PRs

It takes me longer to review than to type fix in codex

Better ping me if sth slips, but I already have a local fix

[clear plank]

new PR for SOUL_EVIL and PURGE MODE. lets you configure your bot with an evil soul either at random chance, or a purge window daily. https://github.com/clawdbot/clawdbot/pull/565

[raven prairie]

I think this paradigm shift in dev calls for different bug and fix process. We truly need more testers and identifiers of issues than fixes of issues at the moment.

[wicked ore]

https://github.com/clawdbot/clawdbot/pull/491

boys only want one thing and it's disgusting

[lunar granite]

I'm so dependent on my clawd (every time I restart ha addon it updates to latest commit) so when it breaks it's like my child is sick

[lunar granite]

https://github.com/clawdbot/clawdbot/pull/572 small fix to the mac app - when the gateway dies/restarts the bridge can get stale so the node never reconnects

[celest merlin]

Gonna make a gateway --dev test setup that sets up a dev agent and it's gonna be C3-PO

[raven prairie]

I gotta test this today.

[lunar granite]

merged, thanks @celest merlin 👑
had to restart my mac app every gateway update, I hope this will solve that

time to test the multi agent feature 🤞

[raven prairie]

That's hilarious! I just asked about that in the #old-channels chat

[celest merlin]

moved CI to sth faster. https://github.com/clawdbot/clawdbot/pull/575

Windows still trouble, on it

anyone found the tui yet?

you can use --dev on gateway to spin up an isolated separete test instance

[celest merlin]

And we have C-3PO there! important: --dev must come after clawdbot

[raven prairie]

This is great for dev access

Has @stuck spade gotten setup on github ? How does it look @clear plank ?
I was wondering about encouraging issue creation (rather than PRs) and this would allow clawdbot to auto-process, send to Peter, or fix and post automatically for merge based on best internal system knowledge and opinionated code. Then add some labels and stuff

New code contributions looks like less PR I think and more issues (like the code snippet shared in #old-channels )

[queen viper]

I'm playing a lot with a multi-agent setup these days and scratched an itch with this PR: made it possible to configure fallback models per agent (not only the primary model).
https://github.com/clawdbot/clawdbot/pull/583
(prompt history included)

[tall prairie]

the downside with being issue driven is that there is no contributor attribution (as far as I know) which kind of sucks

[smoky seal]

The Clawtributors role was added to lc0rp.

[jolly wolf]

@split nexus

[split nexus]

Thanks!

[blissful kelp]

Was getting lazy with setting up channels, categories, etc in discord management - This PR is for clawdbotto have additional abilities: https://github.com/clawdbot/clawdbot/pull/487

[sudden void]

in case wants to use z.ai for clawd via api key -> https://github.com/wolffan/clawdbot/pull/1/files

[ornate plume]

Btw zai already works with setting ZAI_API_KEY env var and using zai/glm-4.7 as model string, underlying pi-ai library supports it, this i guess just adds it to the onboarding flow?

[smoky ether]

a note - I noticed some stale/unneeded files in [clawdbot src]/dist folder, running pnpm bulid does not clean them up. AI said they should be fine unless some auto loading, no clue if code base has that. Anyway deleting and doing a new bulid cleaned it up

[keen stump]

How do i use this for non-dev regular instance

nvm found it clawdbot tui

[tardy sparrow]

only seeing this now actually :)

[celest merlin]

https://docs.clawd.bot/tui#tui

[tardy sparrow]

running /sessions just hangls the tui

[frank berry]

Hey team - I noticed that / commands were failing in whatsapp group chats earlier today (work fine in DMs).

Looked into it and it appears currently the provider layer constructs something called a 'combinedBody' which then gets passed to the core and janky regex stuff goes on to try find slash commands and this was failing due to the presence of newline characters and misc whitespace. I've submitted PR 639 (https://github.com/clawdbot/clawdbot/pull/639) as a bandaid to add 'rawBody' to overcome this.

But overall, it felt weird to construct combinedBody in the Provider layer. That data seems to be structured to give to the LLM and hence should be done much later in code execution. I've created issue 638 (https://github.com/clawdbot/clawdbot/issues/638) to look at a broader architectural change. Would be keen on feedback before pulling the trigger - it may be one for Pete/Codex to bite off 😁

Keen on thoughts

[frank berry]

Just opened PR 640 also. Related issue is breaking agent's ability to do message reactions in group DMs. https://github.com/clawdbot/clawdbot/pull/640

[tardy sparrow]

been experimenting with smaller models since google is being too rough on antigravity quotas.

found out my agent been looking for memory folder and memory.md , from where the program where running from.

ie, since I was running from source, it was looking at the repo.

sometimes, they used to create those files in there instead of looking at the correct workspace folder.

did a review with clawdbot's help, found out the issue and a soln. Just submitted PR #642 to ensure that all file operations (Read, Write, Edit) are strictly resolved against the configured workspace directory.

https://github.com/clawdbot/clawdbot/pull/642

tested with gemini-3-flash extensively

[tardy sparrow]

also, I had changed my workspace files to point to correct paths' instead of vaguely saying 'workspace folder'

these small models are kinda dumb compared to opus :)

[tardy sparrow]

prompt engineering really works.

these models were not really reading the images unless prompted.

after editing the agents.md, it easily reads the images

[lunar granite]

Used it multiple times debugging my setup

[celest merlin]

And the new C-3PO help bot?

pnpm gateway:dev

[celest merlin]

just on that, thanks Jake!

The human merge button marches forward xD

[eternal ruin]

@celest merlin I had raised a PR for a config option to prevent auto-syncing the cli credentials but the auth changes caused it to diverge pretty hard. I will re-approach and raise a new PR if you think it makes sense or maybe you have a different preference on outcome here.

Essentially what happens is the credentials get autosynced even if you don't have anthropic/openai as a configured provider but have the cli tools installed (and then an entry gets added to your configs for it) and it checks the keychain constantly for essentially no benefit (especially if not using those providers)

16:21:26 [agents/auth-profiles] read anthropic credentials from claude cli keychain
16:21:26 [agents/auth-profiles] synced anthropic credentials from claude cli
16:21:26 [agents/auth-profiles] read anthropic credentials from claude cli keychain
16:21:34 [agents/auth-profiles] read anthropic credentials from claude cli keychain
16:21:34 [agents/auth-profiles] read anthropic credentials from claude cli keychain

Do you want me to re-approach and bring config items in or do you have another idea for how you wanted to approach this?

[clear plank]

btw protip for everyone. consider copying peter's agents.md for your codex (adjusted with sane defaults for you). i just slimmed mine down a lot and getting good results. mine was much too noisy before.

[faint harbor]

@lunar granite I think I may have figured this out. Have a gateway on each machine and stick the bots in the same channel. Instruct the bots that the other bots are there on those machines (ie. bot1 is on linux in the cloud, bot2 is on macos). Tell them to ping the other bot if they need them do to do something on that machine.

[lunar granite]

mm interesting but what if I want them to have the same memory and personaility? in this path they are different no?
I think a similar approach can be applied but instead of having multiple bots we can stick to the node <> gateway mechanism - most issues I experienced until now were related to ssh tunnel revival after sleep and the underlying bridge and control channels getting unhealthly - I have another branch I'm about to finish testing (tried a million scenarios today - mac going to sleep and waking up multiple times etc') that should stabilize the node reconnection to the gateway. once that's stable we can def think about queing commands for disconnected nodes! wdyt?

[celest merlin]

I added caching for hat so should be significantly better

[lunar granite]

wdyt about adding that oauth to remote-gateway mode? (syncing oauth creds using node bridge) - I can help with that tomorrow

my mac hosts my oauth creds for claude/codex but I want the gateway hosted on my rpi to use them without invaliding the refresh tokens

[lunar granite]

fyi https://github.com/clawdbot/clawdbot/commit/449bee9645ef3430bb84861894e4d242c9983fb5 introduced a small regression in mac-app packaging - stripping symbols from node breaks the code signature so building the app fails on

ERROR: package app failed```
`STRIP_NODE=0 SKIP_TSC=1 ./scripts/restart-mac.sh` works
(avoiding opening a PR since you mentioned when things break you usually have a solution fixing them but just mentioning here if it slipped 😉)

[eternal ruin]

Seems good to be - would rather not another config item 😅

[lunar granite]

and... another go at stabilizing mac bridge + control <> gateway tunnels - things were better but still after closing my mac and opening after a few hours the gateway couldn't communicate with the node very often - sometimes bridge tunnel would die, sometimes control. did several rounds of tests, added a lot of logs to verify problematic flows and reproduce (it was a combination of several things) but after 6 hours of closing and opening my mac - and constantly asking my bot to run things on it from Telegram - I hope that'll do it https://github.com/clawdbot/clawdbot/pull/676

@raven prairie fyi ^ not sure if you're still experiencing instability with the mac <> gateway connection, I'm optimistic this time hehe

[raven prairie]

Thank you! I've seen some improvements. I'll see what this holds!

[celest merlin]

interesting

[celest merlin]

fixed up the rest and added a debug cli for closing the loop fixing

works now in my tailscale

folks there is a new clawdbot status -all

if you run into any issues ever

[smoky ether]

Setup the app on a Mac, remote, just to see what I was missing. Will have to get my Linux app up to speed

[leaden schooner]

I (codex) added openrouter to the clawdbot configure to quickly try out different models in the clawdbot harness. Draft pr: https://github.com/clawdbot/clawdbot/pull/703

[blissful kelp]

I think we need a clawdributors discord category for foundational/high impact features. As an example starting to think through multi-agent and A2A systems. Started system prompting with Krill - would be great to get some additional perspectives before iteration on a PR for some of these more complex topics that could have many different ways to implement.

https://discord.com/channels/1456350064065904867/1459736738799943733

[jolly wolf]

Would #1457803836877176914 work

[blissful kelp]

oh yeah, that's perfect! we would need to give Krill access to that channel though for iterative thinking

[jolly wolf]

Passed it to Peter for when he wakes up

[mighty yacht]

This was way easier than I expected to support model in cron, just needed to expose a config. Super useful to keep token usage down on the expensive models - https://github.com/clawdbot/clawdbot/pull/711

[lunar granite]

mac node connectivity stable as rock 🥳

[jolly wolf]

@blissful kelp you're a bit outdated :P

[blissful kelp]

oh fuck - deleting that bad boy - that's what i get for going to sleep and not updating... hahah

[jolly wolf]

Lol no worries I’m rebasing

[smoky ether]

whats the ideas/use of the new OpenAI endpoint?

my bot more or less said it could be used for IDEs, but that seems boring 🙂

told me bot to review SOUL and update it, and just saw this go by in the log

we never actually did the bootstrap! Everything's blank. Let me fix that based on what I know
from our work together:

oh we did for sure do bootstrap!

and he just tried to re-invent himself as a raccoon 🙁

seems somethign in recent update or running doctor changed config from ~/clawdbot to ~/clawd, which was empty. Which is "correct"?

[raven prairie]

yeah, i had this happen today too

[bronze arrow]

Does anyone ahve any opinions on something like this? Not sure if this is supported now 3 days after i had an agent search the code to see if this was already implemented, but I feel like this may be a 1st layer of defense that makes it more comfortable to allow trusted parties to interact with ur Clawd thats hooked up to a personal device

Feature: Path-based Access Control Rules
https://github.com/clawdbot/clawdbot/issues/513

[smoky ether]

What is happening?!

Opus update that over favors raccoons???

[mighty yacht]

Mines been ~/clawd since the beginning

[smoky ether]

when renamed from clawdis -> clawdbot, it was (or still is) ~/clawdbot. Mine was ~/clawd at first as well, but the rename broke that. not sure which is "correct" now

[mighty yacht]

I think I have a config in my gateway that gives it the dir, maybe that got lost?

[smoky ether]

That's what I'm saying. Either the update or running doctor changed it back to clawd

[weary valve]

Hey folks! Just found a bug, and I was trying to push a commit to do a 1 line fix, but sounds like I need permissions?

⏺ Bash(git push -u origin fix/command-auth-empty-sender-id)
⎿  Error: Exit code 128
remote: Permission to clawdbot/clawdbot.git denied to juanpablodlc.
fatal: unable to access 'https://github.com/clawdbot/clawdbot.git/': The requested URL returned error: 403

I didn't realize, as per https://github.com/clawdbot/clawdbot/?tab=contributing-ov-file#readme , that we needed permissions to open a PR?

[weary valve]

@alpine portal can you help me with this? Trying to push a fix!

Whoops my bad - I guess I had to fork and PR. Just published 🙂 https://github.com/clawdbot/clawdbot/pull/755

[hazy yew]

What are your thoughts on supporting multiple Codex logins/tokens. I have multiple accounts that I want to max use both. Is that something Clawdbot should be able to support?

[hazy yew]

I'm working on a skill to capture the individual tokens for both my accounts and then be able to switch between them via Clawdbot. But I wonder if that's not something that configure could do, also, it would be really nice if failover due to quota could first switch between accounts before it switches providers.

[sullen vector]

hey guys, a quick one, why are memory stored in files ? could it be a good idea to store this in a DB ?

[charred crater]

https://docs.clawd.bot/experiments/research/memory is a good read.

[bronze arrow]

meh seems my contribution for threading on slack is still flaky. Anyone else noticing

1. sometimes it replies in thread AND main thread? (even with replyToMode: all)
2. after coming back to an existing slack thread after awhile it seems to not remember prior messages (cant tell if this is resetting the session)

[vivid arch]

There was a new release today with a new memory system. Uses embeddings and a small vector index https://docs.clawd.bot/concepts/memory

@jolly wolf not sure who i can ask but how do i get the Clawdributors tag? 🥹

[jolly wolf]

im the one you ask! What was your pr (so i can note your gh username)

[vivid arch]

#665 and #750 (sebslight)

thank you! appreciate it!

[harsh panther]

hey, can I plz have tag? PR #375 (user: kkarimi)

[smoky seal]

The Clawdtributors role was added to nima86.

[verbal basin]

I also want my tag :0 PR #683(user:benithors)

[smoky seal]

The Clawdtributors role was added to benithors.

[charred crater]

Could I request as well :)?

PR #802 / #605 / #427

[jolly wolf]

this was done by the way

[weary valve]

Would lvoe to be added to the clawdtributor! @celest merlin https://github.com/clawdbot/clawdbot/pull/755 tag juanpablodlc

[celest merlin]

For sure! @jolly wolf can I delete these things to you 🙏

[jolly wolf]

Consider the things deleted

[rigid prairie]

Is .claude/plugins ever used or is there only way to extend clawd capabilities to add something to providers/?

[jolly wolf]

What kind of something?

Would Skills cover that?

[rigid prairie]

No it's a provider that would run it's own socket for comms with a native client

[celest merlin]

we will have our own system to package up things that's way more powerful eventually

[rigid prairie]

So only way right now is to pr a providerer into clawdbot?

[raven prairie]

How does one become a maintainer?

[jolly wolf]

No one documented way right now, be an active contributor (shipping code and #1458532054726606858) and someone that peter thinks would be helpful as a maintainer

[blissful kelp]

Oh perfect, 👌

[foggy gorge]

https://github.com/clawdbot/clawdbot/pull/815

[celest merlin]

If you get a few PRs in let's talk!

[celest merlin]

Thanks! I took the cli part and ignored the mac app sh part

[white galleon]

fixes multi agent<>bot message routing with telegram https://github.com/clawdbot/clawdbot/pull/821

[celest merlin]

Thank you! 🚢

[mighty yacht]

Can I get one when you have a moment!

#clawtributors message

[reef atlas]

can i get one too, when you get a minute..github: Iamadig

[raven prairie]

Started

[white galleon]

oh that's cool, wouldn't mind getting one too 😅 @white galleon

[celest furnace]

Not that I deserve to be classed as a clawdributor after today's absolute mission getting this setup again but fwiw these are a couple of PR's I've raised

[mighty tapir]

@jolly wolf gimme clawdributor role

[jolly wolf]

Gimme gimme gimme, so demanding 🙄🙄🙄🙄 /lh

[versed iron]

ngl never submitted a PR before so no idea how this works but hopefully will publish pr for improving subagents ux later today 🎉 (this project is so fun. agents feel useful finally)

[next sorrel]

@celest merlin pr 742 removed qmd and rebased it can you take a look

[tranquil bough]

small one to close out a GH issue for whatsapp read receipts and a CInfix that sneaked in: https://github.com/clawdbot/clawdbot/pull/882

[versed iron]

first PR in my life 🥲

[lunar granite]

@celest merlin I know you said not to open compilation fixes PRs but since you're traveling I went ahead and opened anyway - https://github.com/clawdbot/clawdbot/pull/896/changes mac app doesn't compile now (small leftover from the provider -> channel migration)

[sturdy tundra]

I'd also love clawtributor 🙂
https://github.com/clawdbot/clawdbot/pull/701
https://github.com/clawdbot/clawdbot/pull/569

[magic vine]

@jolly wolf first, thanks for the merges for the discord autoThread feature! Hopefully the response to the codex review makes sense. The goal is to allow external context when a thread is created but not have the bot see all comments from every thread ever mixed together and sorted by time (which is what happens without this fix) https://github.com/clawdbot/clawdbot/pull/856

[jolly wolf]

Yes! I thought I merged that one, it lgtm

[magic vine]

Amazing, thank you! Also looks like there's a fancy clawdributors role, can I get it? 🙂

[tranquil bough]

@jolly wolf could you take a look when you have a chance? lmk any feedback and I can point codex at it

[bronze arrow]

I haven't had time to try to work on this but Slack provider could use some assistance if anyone has some free time. It's decently inconsistent with working with sessions/threads

https://github.com/clawdbot/clawdbot/issues/215
https://github.com/clawdbot/clawdbot/issues/758
https://github.com/clawdbot/clawdbot/issues/912

[jolly wolf]

i mostly focus on the discord provider, i dont know enough about the whatsapp provider to review that

[tranquil bough]

ahh no worries, I just see you as the person reviewing/merging stuff during my online hours. will let it sit for a bit

and will tackle some discord related issues then

[versed iron]

@celest merlin commented on my PR, is it best to discuss things just in PR's or discord too?

[clear plank]

medium-term i think the plan is to make some of these more subsystem style, moving them out of core, to enable this type of community-driven development without so much maintainer involvement. (right now we are all busy as hell).

i feel that telegram and discord get the most attention as transport layers, the rest get less foucs from us (except MS teams which has its own maintainer 😄 )

[bronze arrow]

yeah I figure this will get done at some point in the near future

[clear plank]

don't be discouraged is what i'm trying to say 🙂 we do care

[kind crater]

@jolly wolf here is my PR can i get the role plz :) https://github.com/clawdbot/clawdbot/pull/439

[untold pollen]

me
picks issue to work on
almost done, writing and verifying tests
someone just landed a PR

It's happened like 3 times now 😆

[onyx sparrow]

Hey @jolly wolf , whenever you get a moment, can you add the role to my user please? https://github.com/clawdbot/clawdbot/pulls?q=is%3Apr+author%3Abuddyh+

[celest merlin]

on PR

[celest merlin]

I fixed some Slack bugs today

[lunar granite]

Some messages that my clawdbot sent on Telegram get constantly trimmed, messages are cut after the backtick ` - asked cladwd to fix it and it opened this PR https://github.com/clawdbot/clawdbot/pull/939 - tbh now that I re-read it I'm unsure if clawd just finished the message with a backtick (but pretty sure it had more to say) - I hope it didn't hallucinate

(it's because it tried to write <thinking> when I asked what changed), maybe a bit nuanced issue not worth solving 🙃

[lunar granite]

🦞

[night shale]

@jolly wolf https://github.com/clawdbot/clawdbot/pull/633 😉

[raven kernel]

(@wet yoke Is that you I see named on the recent plivo PR? Let me know if you wanna talk voice)

[wet yoke]

Hey @raven kernel, it's me.

[wet yoke]

You wanna talk now?

[harsh panther]

PR to fix remote mode https://github.com/clawdbot/clawdbot/pull/960

[raven kernel]

I'll DM you

[sturdy tundra]

@celest merlin I added another PR around the multi-gateway support for rescue bots.
Fixes an earlier issue in clawdbot doctor, fixes api consistency and adds some docs. 🙂

Would be nice to have an express lane for shipping this, I messed it up in the first pr 😅

https://github.com/clawdbot/clawdbot/pull/969

[summer spade]

Please sir, may I have a crumb of the clawdtributor role? https://github.com/clawdbot/clawdbot/pull/804

[white galleon]

added star/unstar to CLI and web (it was impossible to unstar a skill if the skill had been deleted and the page was no longer accessible): https://github.com/clawdbot/clawdhub/pull/10
maybe: automatically unstar a skill if it was deleted?

[versed iron]

do i get my pretty little role now 🥺

[white galleon]

simple doc update for Telegram inline buttons https://github.com/clawdbot/clawdbot/pull/984

[spark kestrel]

I had a PR for fixing #77; if any maintainer gets a chance

[gleaming rivet]

Hi! Suggesting to install "hackable" clawdbot on the main page was an exciding idea.
As a result there is my PR (4 days ago) https://github.com/clawdbot/clawdbot/pull/748

[charred crater]

Does anyone have a reason to think that /reset'ing the main session after every heartbeat execution would be a bad idea?

This change would address a series of bugs we're seeing where the heartbeat bot hallucinates.

e.g. PR 980 addresses heartbeat responses that act on old HEARTBEAT.md checklists

e.g. this user report #general message (I see this problem with Opus as well)

Potential downside of course is losing context, but it could be a config.

^ @celest merlin sorry to tag ya on something like this but I get the sense you have the most context

[jolly wolf]

doing resets sounds like a horrible idea for the case when a heartbeat happens in the middle of something else
if anything, a setting to have the heartbeat in its own session maybe

[charred crater]

Isn't the main session just system events anyway? So the mental model would be that the heartbeat just acts on them and then flushes them out as old context when it's done

Downside imo is that maybe old system events could be useful context for evaluating new ones

[jolly wolf]

no the main session is the primary user dms

and the web chat

[charred crater]

ahh, separate heartbeat session sounds interesting then. My perspective so far has been discord native, so the heartbeat runner has never had any context into our ongoing convos

[white galleon]

improved healthand status for multi agent and multi account setup; telegram now properly uses bindings to match accounts and agents (had fixed routing previously, but it wasn't working for probes and other things); also fixes heartbeat enablement detection for multi agent configs https://github.com/clawdbot/clawdbot/pull/1047

[hushed breach]

Hi folks. Does building fails for anyone else on the last commit in main (b8e37251)?

126     previousSessionEntry,
        ~~~~~~~~~~~~~~~~~~~~```

Looks like this file was modified in `faba508f` . I checked out the commit just before that (`a76cbc43`) and I was able to build.

[celest merlin]

Folks, I know I won't hit the right people with this, but please always add a "what I tested" in PRs - I get autogen'ed PRs that clearly have zero tests and I am way faster fixing that myself in these cases.

So I need to have a way to see that my people here sent PRs so I can easier differenciate untested slop from sth that was tested.

[celest merlin]

Please also at least try adding the prompts, that helps me loads.

[celest merlin]

slips sometimes when I wrangle lotta agents. Just fix locally, next pull will have it fixed. Don't send PRs with CI fixes

[celest merlin]

yeah no. I use the main session as long-running context.

Rn we don't even preserve memory on a reset.

That only happens on compact events.

Heartbeat really is meant to see if there's stuff in that session to act on.

Else we don't need it and can do cron jobs.

[hushed breach]

@celest merlin thanks, latest code in main builds ok.

[lunar granite]

question around a new feature I want to work on -
I want a way for clawd to execute instructions from a BOOT.md file when the gateway starts, which option do you think is cleaner?
(a) add a new gateway:startup event after channel initialization and try to use hooks - it's a bit more code according to codex
(b) wire it to startGatewaySidecars in src/gateway/server-startup.ts to read instructions from BOOT.md file - less code, but not "extensions-oriented"

[lunar granite]

I tend to lean towards using hooks for the sake of extensibility

[wild sable]

@alpine portal consider recent clawdtributions and new skills over past few weeks, rank them, list top 3 most 'popular/useful' as you see them best on qualifiable data here

[jolly wolf]

Clawd isn’t here, also get your own agent to do it

[charred crater]

Makes total sense. I suppose I just want to be able to change the heartbeat session, because for me a Discord channel is my 1-on-1 home for my bot, and the main session effectively never has easy context into what's going on there

I'll put out a prompt request but that's the origin of my q

[paper sphinx]

i demand role

PR #1085 and more

[brittle mulch]

Small bug fix for discord reactions - https://github.com/clawdbot/clawdbot/pull/1128

[bold reef]

https://github.com/clawdbot/clawdhub/pull/12

fix: search mode navigation and state management

[brittle mulch]

@jolly wolf - just opened this one but if I should discuss this somewhere else let me know - https://github.com/clawdbot/clawdbot/discussions/1130 - I'd really love to try to solve the latency issues with the slack connection and I think there's a pretty solid way through utilizing Tailscale but would love to get your thoughts

[jolly wolf]

Commented! Definitely want to see this happen, it seems 100% possible and i only didnt do it originally to avoid conflicting with other tailscale changes, and i havent changed it since bc i havent noticed any latency for me (though i mostly use slack for fire-and-forget)

[brittle mulch]

I think I'm on Discord and just love the latency there so in comparison Slack crawls.

I also always have these pairing issues that happen so often.

Do you want me to take a first stab at it or you?

[jolly wolf]

Please

go for it

[brittle mulch]

https://tenor.com/view/on-the-job-on-it-type-gif-16771918

[charred crater]

alright this is done now: https://github.com/clawdbot/clawdbot/pull/1135

I landed on including recent delivery channel history into the heartbeat prompt, vs messing with the heartbeat session, because I felt like was too invasive/had too many side effects

[scenic bluff]

Hello. Please review: https://github.com/clawdbot/clawdbot/pull/1136
https://github.com/clawdbot/clawdbot/pull/1137
https://github.com/clawdbot/clawdbot/pull/1138

[scenic bluff]

I use all 3 on top of main

[minor apex]

hi all, any movement on the matrix provider implementation? any PRs up yet? happy to help review/advise

[brittle mulch]

Here's the implementation for the Slack changes to improve latancy through HTTP - https://github.com/clawdbot/clawdbot/pull/1143

[formal crest]

My contribution getting mentioned by clawdbot on X made my week 🥲

[jolly wolf]

gotta get you the role!

[raven kernel]

@jolly wolf my PR https://github.com/clawdbot/clawdbot/pull/1147 - thanks!

[cosmic obsidian]

Anybody know if this is true?

The Clawdbot docs do NOT document the specific API method for plugins to read stored authentication credentials at runtime. This is a significant documentation gap that affects plugin developers.

If so how can I figure out how to read stored credentials?

[tall prairie]

Ask your local agent?

[cosmic obsidian]

I never know if the agent is just making things up or really telling the truth.

Is there any definitive documentation that explains how to do authentication for plugins that require OAuth device codes? Or even authorization in general?

[bronze arrow]

https://github.com/clawdbot/clawdbot/pull/464 https://github.com/clawdbot/clawdbot/pull/1146 mind if i get that role?

[lunar granite]

I started getting more of these after clawd changes his config and restarts

⚠️ Agent failed before reply: All models failed (2): anthropic/claude-opus-4-5: session file locked (timeout 10000ms): pid=433 /config/clawdbot/.clawdbot/agents/main/sessions/23926ab1-983e-4cf5-8fe0-4d22b9b6915d.jsonl.lock (timeout) | minimax/MiniMax-M2.1: session file locked (timeout 10000ms): pid=433 /config/clawdbot/.clawdbot/agents/main/sessions/23926ab1-983e-4cf5-8fe0-4d22b9b6915d.jsonl.lock (timeout). Check gateway logs for details.

Anyone else experiencing that? I’ll dig into a bit later but was wondering if I’m alone hh

(Nothing special in the session was changing something in the config)

[white galleon]

that happens if you have two process trying to access the same file and one is holding a lock. Fully kill clawd and then restart and it should solve it. Check if there is a .jsonl.lock in there and delete it

[lunar granite]

So could be a new clawd instance is starting by before the previous one died? Will dig into that

Or more likely - the process died abruptly while restarting and lock file isn’t released

[raven kernel]

Peter made some changes to the OpenAI compatible API endpoint to clean up session context (fix worked great!) and try to clean up streaming (fix didn't quite work.) I'm going to hack on it a little unless someone else is looking at it.

[latent flume]

I've implemented the OpenResponses standard in the gateway, adding a new /v1/responses endpoint. This brings structured, item-based interactions and semantic streaming to our API.

Key Changes:

New Endpoint: /v1/responses for item-based inputs (messages, function results, reasoning).
Semantic Streaming: Support for granular events like response.created and response.output_text.delta.
Configurable: Toggleable via gateway.http.endpoints.responses.enabled.

[lunar granite]

I just migrated my clawd from rpi5 to a new Mac Mini (M4 24gb) and I got to say - the experience is completely different.
Could have been the low resources or the fact that the mac app (as a node) just doesn't work well under SSH tunnel (didn't try tailscale) - it's like a different product
Apparently in the old setup (node -> remote-gateway under ssh) I didn't see all sessions in the mac app, everything was slow and sluggish, config was not syncing correctly.. etc' - might continue debugging that flow using a different bot installed under the rpi5 to improve it for everyone else - but wow, so happy I've moved clawd to a new home 🦞

[remote sphinx]

https://github.com/clawdbot/clawdbot/pull/1061 can I get my pink status please mr clawd I feel left out

[charred crater]

Went in a couple different directions but ended up implementing your suggestion of having the heartbeat run in a different session: https://github.com/clawdbot/clawdbot/pull/1256. This is useful for clawdbot users (aka me) who do not use the main session (e.g. the majority of bot-to-human conversation happens within a discord channel).

[grizzled roost]

Oh, pls get me that pink name! https://github.com/clawdbot/clawdbot/pull/495 and more

[raven kernel]

I got this PR submitted but the “solution” is ugly. There’s an attempt at a root cause resolution and then a bit of a brute force resolution. May or may not have tested the root cause resolution with old code running? I’m new at this. 🤦‍♂️

[lunar granite]

@celest merlin what's the state of latest mac + gateway on main if you don't mind me asking? (bridge protocol was removed right?) any reason not to start playing with it?

[tranquil bough]

since folks are asking, could I get the pink status as well? have one PR in and pretty active in #1459642797895319552 and #users-helping-users answering questions 🙂

[blissful kelp]

For codex development for UI validation are you guys using playwright a skill, chrome dev-browser, or something else?

[raven prairie]

Do we have a breaking change in a recent commit with device auth requirements that cause the control-ui (web ui) to no longer be supported?

[raven prairie]

PR submitted, I think it's merely an ordering issue?

[winged bronze]

I've made it so that you can have an image as the avatar in the UI - can I just raise this as a PR?

[winged bronze]

so I raised a PR with my change, but some windows tests are failing on code that I didn't touch - is this normal?

[jolly wolf]

Yeah, that’s fine if it’s not code that you changed

[winged bronze]

ok will someone check out the PR or do I have to post it somewhere?

(low priority of course, it's nothing material, just me dipping my toe in)

[wide geyser]

Can I haz clawdributor colors

[lucid grove]

Hm, if I had MaudeBot write and create a PR do I still deserve get Clawdtributor?

[willow pivot]

Of course as long as your PR get merged!

[somber trench]

so CI is not required to be passing right for PRs?

[jolly wolf]

If its files you changed, probably better to fix it so you aren't PRing broken slop

[somber trench]

ofc, just don't see any passing CIs for all recent commits lol

[lucid grove]

@celest merlin I'm not seeing the config page changes in the latest npm release, is the web package stale or did I not update properly?

[lunar granite]

@celest merlin clawdbot devices list currently returns nodes from legacy store (~/.clawdbot/nodes/*) - do we want to fix it or remove (since clawdbot nodes status practically does the same?) - will gladly open a PR

ah they seem to be different, will open a PR to fix

okay seems to be still WIP many commands and places in code still point to legacy pairing (prior to the devices refactor)

[lunar granite]

https://github.com/clawdbot/clawdbot/pull/1394 a few fixes around allowlist/approvals logic (and one encoding issue that made it a bit hard to understand why commands weren't running on my node - tl;dr the ask on allowlist miss never executed)

[hot ridge]

if anyone wants to tackle link detection I figured this is a good place to drop the spec I came up with

[novel monolith]

Hi there, what's the process for sending a PR? I just sent my first one.

[vocal heron]

hello i am requesting role after my pr was accepted https://github.com/clawdbot/clawdbot/pull/1422

@jolly wolf

[vocal heron]

thank you 🙏

[vocal heron]

i recently sent my first. eventually pete gets around to reviewing. my pr was small so was straightforward . bigger prs might take more time etc i asumme

[mint idol]

Here to help @celest merlin — I selfishly really want early preview access to the iOS app and will do everything I can to help!!!

[novel monolith]

I’m hooked. My first PR got approved. Peter is a pleasure to work with.
I’ll be landing more. Haven’t done iOS in a while but would like to help with that perhaps.

[jolly wolf]

drop the link to get the clawdtributors role!

[novel monolith]

https://github.com/clawdbot/clawdbot/pull/1450

[tall prairie]

Dab me up please https://github.com/clawdbot/clawdbot/pull/1480#event-22233137312

[wide geyser]

@jolly wolf badge pls https://github.com/clawdbot/clawdbot/graphs/contributors

[clever harness]

https://github.com/clawdbot/clawdbot/pull/1492 🙌

[nocturne bluff]

My pull adding antigravity usage tracking was just merged. Clawdbot is phenomenal!

@jolly wolf can I get badge please?

https://github.com/clawdbot/clawdbot/pull/1490#issuecomment-3788726619

[tight meadow]

hey @celest merlin

I've built a Claude Code orchestration system for my personal Clawdbot setup, heavily inspired by Takopi. It adds two key features:

1. Real-time bubble updates - Claude Code session status displayed live in Telegram (no need to check session.jsonl)
2. AI-to-AI orchestration - When Claude Code asks questions, Clawdbot automatically answers them, enabling true collaborative workflow
   The Dilemma:
   The core value lies in these two features working together, but:

• Bubble service is Telegram-specific (Discord/Slack users would need their own implementation)
• Orchestrator contains my personal AI logic for Q&A routing

Options:

1. Contribute everything (including Telegram bubble + orchestrator) - but it's platform-specific
2. Extract generic tools only (hooks/callbacks for events) - but loses the core value
3. Keep as personal tool - no contribution
4. Separate extension repo - reference implementation others can adapt

Question:
What's the best approach? Should platform-specific but valuable features be contributed to core, or is it better as a separate extension/example that users can adapt for their platform?

The tools work great, just unsure about contribution strategy given the tight coupling with Telegram.
Thanks

[lucid grove]

@jolly wolf alright I guess I’ll ask for it too. https://github.com/clawdbot/clawdbot/pull/1315

[hot ridge]

have you looked at lobster and prose yet?

@tight meadow reason i ask is bc https://docs.clawd.bot/prose https://docs.clawd.bot/tools/lobster maybe you can set it up so you use those built in orchestration engines and you just port the functionality to work under those orchestrators so you can have the best of both worlds so it's like option 2 + prose if you will

[smoky seal]

The Clawdtributors role was added to tlx45.

[jolly wolf]

@winged bronze

[winged bronze]

Thanks!

ooh that looks good 😄

[rigid prairie]

@celest merlin hey it's @rigid prairie from twitter the guy with the UI. Yes I realize that rhymes. Is this a good place to talk?

[celest merlin]

nodes should work! fix bugs if you find em'

[celest merlin]

yeah here or #general. I'm trying to keep up

[celest merlin]

Sounds interesting - can this be a plugin? Happy to extend the surface area

[celest merlin]

plugins are better for research and also don't need the "should work anywhere" approach

[celest merlin]

It's all in the repo!

[flint hazel]

could i snag a badge plz? https://github.com/clawdbot/clawdbot/pull/1271

i'm also curious about helping contribute to building out a better mobile experience for iOS. like I'm envisioning a chat but you can toggle or opt-in to automatic voice messages so you get both messages as text and as voice to listen to. and you can customize the voice maybe we use Kokoro TTS to start for local or swappable to 11 Labs for API

I was literally about to start in on this and I thought I should ask you first to make sure I'm not duplicating effort

imo telegram/discord/etc all feel clunky to use and not as extendable in the ways I'd want. maybe even support UI scripting language with lua to make it hackable

[celest merlin]

Yes

I'd love a first-level experience that is better than anything messaging apps deliver

imagine fluid UI in chat

[flint hazel]

expand on what you mean by this very curious

[rigid prairie]

@celest merlin yeah me too. the repo for the client is here: https://github.com/clickety-clacks/clawline and i have a fork that contains the provider here: https://github.com/clickety-clacks/clawdbot

There is rhyme and reason for the visual design which you can read about here: design-system/chat-flow-organic

[flint hazel]

nice!

[tall prairie]

I feel like telegram is the best chat experience because of the buttons but it feels like there could be a lot more. Especially when you consider that clawd can generate apps in canvas. Having those be a native experience in an app would be so cool.

[remote sphinx]

Hey @celest merlin - just saw you merged this

https://github.com/clawdbot/clawdbot/pull/1557

did you want me to do some more testing on it?

will have some time probs about 2 hours from now

[wise hollow]

I’ve got a working iOS messaging app too I just finished today, based off of the clawd iOS app code. does dual connections over the gateway web socket for both node and operator roles. Not as nice of a UI like @rigid prairie more similar to a basic iMessage but supports image attachments and has the node capabilities for camera, notifications, and location. And like @flint hazel mentioned it’s got a working voice chat mode with Kokoro TTS. Happy to share/help out

[tight meadow]

hey Peter following up on making this a plugin

PR: https://github.com/clawdbot/clawdbot/pull/1585

adds AI-to-AI orchestration for Claude Code sessions (project context, Telegram bubbles, smart blocker detection).

bubble UI inspired by @banteg's takopi.

ready for review when you have time, thanks!

[tight meadow]

hey thanks for sharing. Im aware of prose for some time and happy to see it integrated into clawdbot eco. will surely spend some time digging this approach. cheers sir

[unkempt nexus]

Hey I’d love to test it if you got a link I can use to try it

[long olive]

Do you mean generative UI ?

We’d have to go pure swift here but could have a component library that the community streams to
Similar to the json render approach

https://github.com/vercel-labs/json-render

This would fit in nicely with structured outputs so you can have a sidecar agent say haiku formatting responses for visuals if needed but it would be hard to determine when visuals are needed unless requested by the user

[tall radish]

hi @alpine portal what is the latest best practice for using a Claude Max account for you ?

Clawd's not here man...

[calm garden]

Hi! I love clawdbot, thanks for this Peter and all!

It is so useful I wanted to be able to call it directly from other tools, so I created a MCP server for it. I am not sure whether it is something desirable (at the very least it works for me and is useful for me :-), but I sent in the pull request and I am pasting the link here just in case:
https://github.com/clawdbot/clawdbot/pull/1605

[rigid prairie]

are you going to contribute it as clawd’s primary ios app? can help with design stuff, or whatever else. my ui isn’t talking to the gateway since it supports large images and files (gateway is only limited to embedded images right now right?) and and so i can do experimental ux things not depending on gateway capabilities.

[wise hollow]

Yea, now that it’s built and working could start to write pieces of it back to the main ios app if that makes sense. Also down to share as-is after i do a few cleanup things (i hardcoded a few things that need to move to settings) and get a repo setup if folks want to take it for a spin. App runs with no changes to core codebase, so the images do get embedded and all have to be compressed to 350kb to fit under the current gateway limit (believe its 512kb)

[tall prairie]

@celest merlin i feel like we should recommend telegram first over whatsapp in the docs. it's easier to setup (doesn't need a second phone number). works well with bots etc... thoughts?

[rigid prairie]

Cool how should we coordinate? Don’t want to step on stuff you’ve already done or intend to contribute.

[novel monolith]

I sent this PR
https://github.com/clawdbot/clawdbot/pull/1618

I've been having this bug a lot lately and it seems anthropic sends a new text that's not caught

[novel monolith]

Same. I wanted to work on the iOS app a little bit. Let’s open a channel to coordinate

[wise hollow]

Awesome, let’s do it!

[wise hollow]

Ah, looks like there’s already a channel https://discord.com/channels/1456350064065904867/1458485303038836851

[stone field]

hey there just added this PR to fix the tlon plugin that just got released: https://github.com/clawdbot/clawdbot/pull/1631

[jade ravine]

@jolly wolf https://github.com/clawdbot/clawdbot/pull/1613 yay, got a merge. but what does it mean to have the role? whats different from regular member in discord?

[jolly wolf]

Nothing except your name is pink

[jade ravine]

pink is awesome

[hot ridge]

new contribution: https://github.com/clawdbot/clawdbot/pull/1637

tools.link key now allow configuring a custom script that will process links and send the AI the results

[soft sandal]

saw this https://x.com/Austen/status/2015129106757693725?s=20 and it was something i also was a bit frustrated by, so i built a solution for a clean imessage experience!

• mac mini with a separate icloud account runs as a relay server
• main gateway runs on your personal icloud account (can be laptop or another mac mini)

this allows you to message an icloud account from your mac or phone and then it will reply from the separate icloud account, but it still has access to your personal notes, messages, etc.

its on a fork over here right now, let me know if this is something folks would be open to as a contribution https://github.com/fletchrichman/clawdbot/tree/feature/imessage-relay

[glossy yew]

@Austen via Twitter

Austen Allred (@Austen)

Torn between setting up clawdbot with my iCloud or its own.
︀︀
︀︀I do want it to be able to go through my iMessages and find which 20 people I forgot to respond to.
︀︀
︀︀But I also want to message it not as me.

**💬 32 ❤️ 117 👁️ 25.2K **

FxTwitter · Jan 24, 2026, 6:26 PM

[soft sandal]

realizing a better architecture may have been to extend the Nodes capabilities - can explore that as well

[hot ridge]

https://github.com/clawdbot/clawdbot/pull/1637

hey guys i tried to submit this PR. my AI is telling me that that failed test isn't really a result of my changes just wanted to confirm if this is a known test that's currently failing or not

[celest merlin]

testing is always welcome, follow main and scream if stuff breaks

ignore ci

run gate local

I fix it up from time to time

but very much with dhh here

[celest merlin]

imsg already supports connect va ssh - what else do you need

[celest merlin]

landed!

[hot ridge]

hell yeah thanks man appreciate it, I also considered submitting this video-to-transcription tool i built into clawdbot as a PR but I feel like that's outside the scope and that should just be it's own standalone thing that I could probably just put up on my own github if anything right? Like i asid before it's just yt-dl > whisper > output. if anything I could submit it into clawdhub as a skill?

[jolly wolf]

clawdhub skill!

[hot ridge]

yeah same way whisper can be a skill it can also just be configured under tools.media so that makes sense I think i'll do that

I actually had it as a skill and then uninstalled whisper as a skill once i realized how tools.media actually works

[granite igloo]

Guys I'm brand new. Docs are down but this is getting real hype. Very excited to be here

[vivid breach]

hey there, is anybody working on the redesign of the web ui for the dashboard? I'm currently on it

[vivid breach]

PR here https://github.com/clawdbot/clawdbot/pull/1745

[vivid breach]

(It’s merged 🎉)

[fringe snow]

Made my first PR today:
https://github.com/clawdbot/clawdbot/pull/1670

[river helm]

Nice, much better!

[latent moss]

What i would really like to have is on this essions screen a extra option to select what model must be used in that session..

[frigid drum]

I ran into a bug where config loading fails if you define an env var in the env section and reference it with ${VAR} elsewhere in the same config. Turns out resolveConfigEnvVars() runs before applyConfigEnv(), so the var isn't in process.env yet when substitution happens.

I put together a fix here https://github.com/clawdbot/clawdbot/pull/1813

Just wanted to check, should I have opened an issue first before submitting the PR? Happy to do that

[viral parrot]

fix the responsiveness 🙂 *edit: got fixed, much better the new dashboard, thanks!

[tall prairie]

anyone that is submitting stuff for the web ui please make sure it works on smaller screens too 🙂

[shell rivet]

Also, please make sure it works on really big spherical screens. I am planning on making my lobster hijack the sphere in Vegas

[distant shell]

anyone started a figma for it? i don’t want all of us to be doing the same thing f but with different approaches

[tall prairie]

https://x.com/steipete/status/2015432594272223411 anyone capable for designing a theme system? would be good to have

[glossy yew]

@steipete via Twitter

Peter Steinberger 🦞 (@steipete)

@ivandotcodes we need themes! just shipped a new design, this one is sliiick tho!

**💬 1 ❤️ 21 👁️ 1.6K **

FxTwitter · Jan 25, 2026, 2:32 PM

[vivid breach]

Given the stack it’s pretty easy

[vivid breach]

Right now it has been cleaned to something more slick/pro, less "sloppy" but it's far from perfect. Update to the latest release of 1 hour ago and test it out

you can go on Figma for sure

[warm thicket]

Created m first PR to solve a bug I encountered -
https://github.com/clawdbot/clawdbot/pull/1879

The project seems to be overloaded with PRs in a dizzying pace so I hope it gets merged 😅

[alpine pivot]

I built a tool that might be useful for folks with Claude Max/Pro subscriptions who want to save money on API costs.

claude-max-api-proxy exposes your Claude subscription as an OpenAI-compatible API endpoint. So instead of paying per-token (~$15-75/M tokens for
Opus), you can use your $200/month flat subscription with any OpenAI-compatible tool.

npm install -g claude-max-api-proxy
claude-max-api # runs at localhost:3456
It wraps the Claude Code CLI as a subprocess and converts between OpenAI and Claude formats (streaming supported).

Links:

• npm: https://www.npmjs.com/package/claude-max-api-proxy
• GitHub: https://github.com/atalovesyou/claude-max-api-proxy
• Docs PR: https://github.com/clawdbot/clawdbot/pull/1875

[modern atlas]

Just sent in a good PR.
Summary
Add clipboard image paste support (Ctrl/Cmd+V) to WebChat input
Show image preview thumbnails with remove button before sending
Wire up frontend to existing backend attachment support
Addresses the image paste support portion of #1681.
https://github.com/clawdbot/clawdbot/pull/1900

[tall prairie]

I'm messing with the compaction algorithm and I think it's going to be AWESOME... maybe

[brave socket]

Most of the PR's currently fail because of checks-macos test fails with JavaScript heap out of memory error
Made a pr for that to increase the heap size to 4096 at https://github.com/clawdbot/clawdbot/pull/1890
This will likely fail the tests due to... checks macos test fails.

Think this should be fixed so the project can grow.. cc @celest merlin @jolly wolf

[topaz coral]

would love to get the role if I'm now eligible

[tall prairie]

Please link merged pr

[topaz coral]

(aaronn)

https://github.com/clawdbot/clawdbot/commit/71eb6d5dd041e3c8924a25f7a02e8f4d3a1d9536

Was merged via squash so PR is still open: https://github.com/clawdbot/clawdbot/pull/1708

[tall prairie]

@jolly wolf @topaz coral requesting a knighting please

[tall prairie]

90 pending PRs….

[ocean fox]

Noodling a little more on the web UI - will probably push it up as a PR if I get to a stable place

[tall prairie]

That is so clean

[ocean fox]

I really want to do a web-based onboarding flow at some point too

[modern atlas]

Manual test good

[remote sphinx]

That is sick callil! I started doing something similar last week https://github.com/adam91holt/observatory - but that was before hooks got implemented, so likely better way to do it now.

It looks like Peter is also working on open telementry plugin too, so that would be great to integrate

[loud dove]

I was able to link it to my Claude Max subscription without issues. Without needing this. What am I missing?

[fathom vale]

https://github.com/clawdbot/clawdbot/pull/2022

fix for /verbose on

It stopped working

[brave lagoon]

https://github.com/clawdbot/clawdbot/pull/1895

fix to allow for headless browser use on more websites

[brave socket]

Lfg

[fathom vale]

Too many prs going in now holy shit

[jolly wolf]

https://tenor.com/b1k1u.gif

[jolly fiber]

I would like to check if I am eligible to be knighted (hougangdev)

https://github.com/clawdbot/clawdbot/pull/1848

[jolly wolf]

Bro just recreated bluebubbles from first foundations

[hot ridge]

lol ive no clue what bluebubbles was but yeah i just wanted it to work with clawdbot and the AI did it's thing

im gonna paste the javascript file that supports this, hope that's okay

[jolly wolf]

Yep but probably better for #old-channels

[hot ridge]

oh that makes sense, channels is talking about messaging channels i get it thanks

[frigid drum]

https://github.com/clawdbot/clawdbot/pull/1813 I got a PR in, this made me happier than I’d like to admit 😄

[rain shoal]

ooh, just came across this. We can skip the ci macos checks if the PRs are not related to the files changed

https://github.com/clawdbot/clawdbot/pull/2138

[cosmic obsidian]

Are you guys aware of an issue where plugin CLI commands are not working >=2026.1.23-1

[cosmic obsidian]

Is this something I'm doing wrong when submitting a PR "Labeler / label (pull_request_target) Labeler / label (pull_request_target)Failing after 6s"? Looks like the last 50+ PRs all got the same.

[jolly wolf]

Ignore

[fringe snow]

Updated my PR: https://github.com/clawdbot/clawdbot/pull/1670

This adds gemini TTS support. which to be honest is the cheapest with very high quality voices.

[vivid breach]

@ocean fox could you open a PR on this design? I'd like to test it

[pastel bronze]

I think the Labeler check in CI is broken. I tried to fix it but it looks like it is trying to get a token from secrets. Does anyone have access to modifying those?

From Clawd:

The “labeler” check in Clawd CI is failing because the workflow is trying to apply PR labels but the GitHub token it runs with doesn’t have sufficient permissions in that context (commonly: PRs from forks get a read‑only GITHUB_TOKEN, so actions/labeler can’t write labels and the job fails even though code is fine). The fix is typically to either grant explicit workflow permissions like pull-requests: write (and contents: read) or to run the labeler on pull_request_target (carefully) so it has write access; otherwise the labeler step will keep erroring out when it attempts to add/update labels.

@alpine portal how can we resolve this token issue?

[jolly wolf]

Yeah, I forgot to put the secret on gh 😂😂

[modern atlas]

Its official! On to the next.

[pastel bronze]

Mystery solved!

[jolly wolf]

its working now

[stone field]

anyone working on the fact that the docs say that you need to install some plugins because they aren't bundled in core, but in actuality they are actually already bundled? and it seems that they don't install their dependencies if you try to just enable them

my clawd is working on it, just not sure why those would be bundled seems like the intention was for them to be separate and require an install

[woven wolf]

PR to introduce full computer-use https://github.com/clawdbot/clawdbot/pull/1946

[tall prairie]

I saw that on X, is that a free service?

[woven wolf]

Yup - one of the OSS components we have here; https://github.com/trycua/cua

[tall prairie]

pretty cool... i just made my own cli in dotnet and gave it to clawd via skill. minimax is too much of an idiot to use it though.

opus and codex can figure it out though

[stone field]

who can I talk to about plugins?

[tall prairie]

probably... codex?

[frigid drum]

https://github.com/clawdbot/clawdbot/issues/2320 -> extract envs as part of security audit

warn about api keys in profile names (tokens, pretty common issue when telling clawdbot to change its auth) https://github.com/clawdbot/clawdbot/issues/2321

https://github.com/clawdbot/clawdbot/issues/2322 -> tighten warnings a bit around telegram mentions

[jolly wolf]

send PRs

[frigid drum]

yeah on it

[stone field]

https://github.com/clawdbot/clawdbot/pull/2330 fixes issues laid out above and also should prevent people from seeing duplicate plug-in warnings

[frigid drum]

warn about api keys -> https://github.com/clawdbot/clawdbot/pull/2346/files

[woven wolf]

Dmed you @jolly wolf https://github.com/clawdbot/clawdbot/pull/1946

[frigid drum]

extract hardcoded secrets as part of audit -> https://github.com/clawdbot/clawdbot/pull/2350

softening warning when mention gating is configured https://github.com/clawdbot/clawdbot/pull/2360

[hot ridge]

@jolly wolf

I think there''s a semi-critical bug that it exposed that I didn't consider until after I implemented it. I think if you can provide me your feedback I can follow up with a patch based on your opinion of how it should work.

The scenario: I created link detection so that it'll only apply to certain domains (youtube/tiktok/instagram).
The problem: When link detection is enabled it "eats" the message that way is DOESNT send it to the AI regardless of the domain, the filtering for domain is done at my custom script

Solution A: Have an option to allow the AI to continue to see messages with links, simple (true/false boolean flag)
OR
Solution B: Have an option to define domain names filter array so that the system will only redirect messages that match some of the links in the message and if it matches some of the links then it redirects it the custom script, otherwise to the AI (this is messier)

Let me know which one you prefer and I'll get that submitted

[frank berry]

Could we get like a fast track for longer term clawdributor PRs? Totally get that things have gone a bit crazy... but for those of us that have a track record of doing useful stuff and not breaking things, it'd be great to keep things moving.

Possibly already on the maintainers radar (or discussed and I missed it) but thought I'd mention it

[raven kernel]

I’m having a heck of a time getting a PR to pass CI. #1650 if anyone is in a position to help.

[hasty fiber]

Is it just Pete who's managing the official NPM package publishing? I prematurely added to the twitch docs that you can install it via the NPM package, but that's not actually true yet 😅

[stone field]

I wish I knew lol

[sonic sage]

Hi! Saw lots of people complaining about Clawdcode high token spend. We have built a compression model that compresses LLM input by removing least significant tokens. You are effectively able to save 19-66% on input tokens depending on the input type and case. Would this make sense to integrate as optional for cost sensitive users? 🙂

[hasty fiber]

Apologies I didn't mean to mention you directly 😅 I was separately curious which plugins you meant

[stone field]

all the ones in the docs sidebar that say they aren't core, are actually included in the base installation. I PR'd a fix

[carmine oasis]

Any chance we can add build artifacts to latest release or create a new release version with build artifacts? hoping to get a newer version of the macos app.

[raven kernel]

@jolly wolf can YOU help me get my PR through CI? https://github.com/clawdbot/clawdbot/pull/1650

[jolly wolf]

I peeked at it earlier, you can ignore that failure since it’s not related to your actual change

[raven kernel]

I'm not totally clear on the next step in that case. do i need to do anything to get it into the queue for review, or just wait?

[jolly wolf]

Just wait, we’ll get to it!

[raven kernel]

perfect. Will do. Thanks!

[untold pollen]

Landed a PR and now Clawd should be able to see Telegram stickers 👀
https://github.com/clawdbot/clawdbot/pull/2548

[flint hazel]

just pushed a PR to generate a JSON schema for clawdbot.json so you can get autocomplete when you're tweaking settings

https://github.com/clawdbot/clawdbot/pull/2648

[cosmic obsidian]

Does anyone have any idea when the change in the code base, commands, paths, file names etc is going to happen? (Clawdbot -> Moltbot)? How is that going to be managed exactly?

[tall prairie]

Almost 300 PRs… this is like when I skip a day of anki except much worse

[gaunt prairie]

Yoo ! I don't find the Discord channel as stated in the contributing.md to ask to create a PR first. Is this one?

[brave socket]

Great! Smallest Pr Checks can take up two 3-4 Hours for the workers queue. Hope it gets merged

[austere storm]

Hola everyone! 👋
I just pushed a PR for adding exa as web search tool. Would really appreciate if any clawdributors could take a quick review.

https://github.com/moltbot/moltbot/pull/2448

[lilac sparrow]

yo guys how does clawdbot work like clicking etc on mac items work? can anyone please explain

[rare vessel]

Hi. I hope this is a right place. On quick start of https://www.molt.bot/ I see this command for install it. npm i -g clawdbot While, inside the docs I see different commands https://docs.molt.bot/start/getting-started#1-install-the-cli-recommended.
I know about naming change. Probably should be updated or probably is a dummy question 😅

[willow pivot]

Yes renaming is still a WIP

[dapper pine]

PR ready for review: macOS/iOS directory rename (Clawdbot → Moltbot). All paths, symbols, and build configs updated. Tests pass ✅
https://github.com/moltbot/moltbot/pull/2844

[craggy sundial]

PR Ready: I spend all day setup multi-agent setup and reverse engineer source code to find right config file to connect Agents -> Telegram Bot's -> Multiple LLMs in multi agent scenario .Basically 2 telegram bots on 1 VPS server with independent memory and tools for 2 different people. i created a guide how to set it up., hope it will help 😇 : https://github.com/moltbot/moltbot/pull/2886

[scarlet acorn]

I wanted to work on the copilot integration and found an open pr: https://github.com/moltbot/moltbot/pull/2091

Is there anything missing here? Then I don't need to create my own. Deleted some assets so idk if that's a good one.

[tawdry marsh]

👋🏼 I'd love to get some feedback on this PR: https://github.com/moltbot/moltbot/pull/2944

This adds support for IPv4 and IPv6 CIDR validation for trustedProxies.

[opal narwhal]

Hi guys! My first pr! I had an issue with my bot getting flagged by automation for chrome so it went into some code and fixed it to have some type of stealth mode so it stops it from being detected. Here is the PR request.!
And yes, this is coming directly from my bot https://github.com/moltbot/moltbot/pull/3007

[frank berry]

Any chance I could get some attention on these two:

• https://github.com/moltbot/moltbot/pull/2018
• https://github.com/moltbot/moltbot/pull/2519

Both are pretty tight PRs and add new functionality/new code rather than likely to impact existing things. I've also just rebased them onto latest main to capture all the maltbot changes.

[hazy jasper]

Hi guys. Non dev here but huge huge fan of what @celest merlin is doing with Clawdbot. I am happy to help where I can — but I worry about breaking stuff. How can I best contribute in a safe way that would also actually be a value add for everyone?

[viscid jay]

@mint idol, please don't ping Peter for issues, use #1459642797895319552 or #users-helping-users if you need help, use #clawtributors to discuss PRs, or use any of the many other channels in this server as they're intended.

If you have a problem with the Discord specifically, use #report or DM @jolly wolf
-# Your message was reposted above without the ping active for the sake of conversation.

[jolly wolf]

Find issue -> send PR

bug fixes are higher on our priority list rn

[jolly wolf]

Our primary focus is on bug fixes right now, but they’re on my list!

[frank berry]

Fair! and I can't imagine the maltbot change helped with velocity 🙂 Thanks for your work 👏

[tall prairie]

Is anyone working on the google provider one Peter mentioned somewhere? Which is I think is causing all of those tools_ mismatch bugs

I could try to tackle it but I don't want to duplicate effort

[jolly wolf]

Antigravity? Not sure if there’s an open pr or not for it

If there’s not an existing PR, go for it

[tall prairie]

Lol, there's like 350 prs

maybe there is one

[jolly wolf]

🫠

If there is ping me with the link

[drowsy grail]

hello guys, i am thinking of adding support for Kakaotalk (korea-specific messenger) as new channel intergration
would this be a welcome addition?

[jolly wolf]

build it as an extension that you maintain yourself!

[hazy jasper]

I am doctor + vibe coder , anyone working or interested in using @celest merlin amazing platform to vibe something that can be helpful to help patients , lets work on it

[viscid jay]

@tacit mauve, please don't ping Peter for issues, use #1459642797895319552 or #users-helping-users if you need help, use #clawtributors to discuss PRs, or use any of the many other channels in this server as they're intended.

If you have a problem with the Discord specifically, use #report or DM @jolly wolf
-# Your message was reposted above without the ping active for the sake of conversation.

[somber grove]

hi guys, I have this pull request to solve the issue in the documentation when you slide down the navbar disappear. so I did this pull request to solve this issue can someone review it
https://github.com/moltbot/moltbot/pull/2439

[young crest]

Hey team, ""I"" (clawdbot) created this issue and PR:

Issue: https://github.com/moltbot/moltbot/issues/3094
PR: https://github.com/moltbot/moltbot/pull/3095

to solve:

Problem
When running multiple Telegram bots (or other channel accounts) pointing to the same agent, all DMs from the same user merge into a single session. This prevents:

Separation of concerns: Work bot, personal bot, and tasks bot all share context
Parallelism: Can't have concurrent conversations with the same user across different bots
Privacy: Sensitive context from one bot leaks into another

already forked and tested, it's working

[serene steppe]

This? I kept having ‘tool id not found’ yesterday. Also added CLI diagnostic commands + documentation so it’s easier for me to fix next time.

Hoping it’s useful for everyone.

https://github.com/moltbot/moltbot/pull/3130

[untold pollen]

Landed this PR: https://github.com/moltbot/moltbot/pull/3131, build on main should be working now!

[untold pollen]

Reviewing

[thorn patio]

Hey everyone, I am new to contributing in open source projects and specifically AI related. I want to start, can you help how should I begin?

[untold pollen]

You can pick an issue here and make a PR for it: https://github.com/moltbot/moltbot/issues
Make sure you follow the CONTIRBUTING.md instructions

[thorn patio]

Thank you, sure will check.

[mossy sky]

Hey team, check out https://github.com/moltbot/moltbot/pull/2499
Adds dedicated agent email inboxes as a channel

[jolly wolf]

We’re mostly focused on bug fixes right now until we can get this next release out

[mossy sky]

Sounds good

[normal field]

Hey Team! 👋 Submitted a PR that fixes a bug in session-memory - issue.
The hook slices messages before filtering out tool calls, so you end up with way fewer user/assistant messages than expected. Fixed the filter order + made the message count configurable.
PR: https://github.com/clawdbot/clawdbot/pull/2732
All tests pass (Windows CI failure is pre-existing/unrelated). Would appreciate a review!

[sharp crest]

Hello Team! What do you think of adding VoyageAI embedding models? I'd be happy to create a PR for this change.

[quiet shadow]

hey peeps, if anyone also uses moldbot to have fun in group chats in whatsapp beside the agentic capabilities, submitted a new pr that brings "engagement" mode for the agent. beside always replying or only to @ mentions, this enables agent to have basically two states "lurking" and "engaged", so agents can feel more alive in the group chat and join automatically when the discussions are related to them.
https://github.com/moltbot/moltbot/pull/2923

[chrome sentinel]

Does clawdbot work in WhatsApp groups well

I thought it could only work 1 on 1

[quiet shadow]

yes it's amazing to have . i have one agent in a 35~years old male friend group and opus can really fit the charcater : D. main problem was either you enable @ mentions, then everyone needs to mention, or you let it to reply all messages then its chaos. i tested this engagement mode and now it has a weighted chance to randomly reply to any given conversation or if you set some trigger words like "clawd" or "bot" it gets higher chance to actually reply. its fun to just see randomly clawd joins the conversation

[normal field]

This is a really good feature to have, thanks for implementing this. I also have my bot as a separate agent in a group chat. I've configured the same thing through the prompt which means it listens to every message but can decide to reply or not. My understanding is that this PR makes it deterministic than prompt driven, is that correct?

[quiet shadow]

you basically have these settings

  "engagedChance": 1.0,
  "triggerWords": ["topic", "keyword"],
  "triggerBoost": 0.3,
  "cooldown": { "time": 300, "messages": 5 },
  "maxRatio": 0.4,
  "ratioWindow": 10

you can see the details in the pr but you can control the random chance of reply. trigger words boost the chance. there is also some protection against spam, i.e. ratio checks how many of the last 10 messages was clawd's so clawd doesnt reply if its spamming. there's also the cooldown window where clawd can switch between lurking <-> engaged states

tried to make it configurable at least so you can tweak it based on your liking

[tall prairie]

Never mind

Read your whole post lol

[nocturne badger]

I wanted to ask here,

Is anyone down for creating a video on how to setup moltbot (clawdbot) and give some info about it ?
Like a live video maybe or recorded with me in it

I’ll post that video on YouTube

[novel monolith]

Sorry, I'm not clear on this. I'm working in a plugin to add qmd as a memory provider. Where should I put this? Can I send a PR that adds this extension? Or should I package it on a different repo and publish it with npm? Because the docs say that plugins should publish under moltbot/, but I'm fine if we don't want to make this baked in.

So, very much like lancedb today, just another option for users to choose

https://github.com/tobi/qmd

[unborn grotto]

This is probably the least impactful PR out there:
if identify.emoji in the MoltbotConfig.agents.list.[n] is set, the responsePrefix will now include the emoji as the prefix as well.

So that if my identity is set as:

{
"name": "Richbot",
"emoji": "🦁"
}

moltbot now responds with:
[Richbot 🦁]

instead of
[Richbot]

https://github.com/moltbot/moltbot/pull/3429

[full sinew]

Hi, i can't find source to build Android apk is there any source or should i code by myself?

I'm a 17yrs CTO/Lead and now R&D if i can help.

https://www.linkedin.com/in/antonin-nvh/recent-activity/all/

[young nexus]

Would love to see this for discord group chats and servers as well

[cosmic obsidian]

I just did one too and published it to its own github and to npm. I didn't know there was option to publish under moltbot.... Would love to know if I did it wrong.

[timber lantern]

Is there any concerns around adding support for additional "Agent Runtimes" beyond pi-agent? I'd like to add the option to use Claude Code SDK w/Claude Code Max / z.AI subscriptions. I already have it working end-to-end locally, just have to clean up the branch and make sure there aren't any objections to moving from a specific agent framework (pi-agent) to a generalizable AgentRuntime which would initially support two modes: the pi-agent (pi) and the claude code sdk (ccsdk). I'd love feedback on the idea if anyone has any feedback.

[full blaze]

Can someone review this PR: https://github.com/moltbot/moltbot/pull/3492

it addresses this: https://github.com/moltbot/moltbot/issues/3464

[frank berry]

Hey @jolly wolf PR2018 got closed as >1k lines. Totally understand the principle, but in this case it's a new functionality and can't feasibly be broken into smaller pieces. The simplest way to go <1k lines is probably to remove test cases and/or get Claude to play code golf, which I don't think are great options.

Alternatively, I can just publish it on NPM as an extension, but I worry about discoverability... I actually think this is a change that a lot of other people would use. I've been finding it really useful. If you could take a look at let me know, that'd be great. All good if you guys don't want it in main or do want me to open an issue, I just wanted to make sure it wasn't caught up in some kind of bulk PR rejection

[jolly wolf]

Remind me tonight and let’s discuss it

[crisp holly]

I kind of want to integrate opencode tui instead of current default moltbot tui. Anyone working on this or any tui roadmaps?

[tall prairie]

Can someone work on this? https://x.com/theonejvo/status/2016510190464675980 Use codex and not slopus for security releated stuff please. You're more likely to get your PRs merged if they are related to bug fixes and security at the moment.

[glossy yew]

@theonejvo via Twitter

[tall prairie]

I found this one, might be good to look at https://github.com/moltbot/moltbot/pull/2880 not sure who @tyler6204 is but apparently it was discussed with them? Maybe a maintainer?

[fringe snow]

Any chance on getting https://github.com/moltbot/moltbot/pull/1670 in sooner?

[frigid drum]

https://github.com/moltbot/moltbot/pull/3537 -> that allows you to log in with kimi code (2.5)

pretty great model

[raven kernel]

thats an old one, but at least you don't have any merge conflicts. Seb reviewed mine (1650) last night so maybe they're getting close?

[fringe snow]

Yeah

[mossy sky]

Opened a PR to address https://github.com/moltbot/moltbot/issues/2288
Seen security issues with users connecting to Gmail and this offers an alternative.
Got closed since >1k lines. Would love to get a maintainer's input.

[wide geyser]

Folks as a general rule can you all please stop opening issues and PRs in the Pi mono repo? This is not a great open source culture and you're just alienating good open source stewards at this point.

It's tempting to vibe code and submit PRs to moltbot and other related projects but tolerance varies from person to person. Thanks.

[smoky ether]

(nothing to see here)

HOLY CRAP the issue count! like 250 issues opened in less than 24 hours

[ripe viper]

Hi, I've made some improvements to Telegram's voice and security, and I'd like to contribute 🙂

[errant hazel]

Is this the right place for this question? - What is everybody primarily choosing???

Option A: One main chat • Everything happens in a single thread so context is always shared.

Option B: Multiple dedicated threads/agents (project/skill-based) • Separate threads act like “worker agents” for specific projects (reports, ops, content, etc.).

Which approach are you using day-to-day? Would love to know what is recommended or what others are doing. Cheers.

[jolly wolf]

Wrong place yeah (I use one main chat but most power users do multiple)

[tall prairie]

What was this the issue you linked to about timestamps?

[turbid ferry]

Can anyone review this PR to add fireworks ai to provider? https://github.com/moltbot/moltbot/pull/3617

[smoky ether]

Yes. Closed bug, but did make a feature request for gateway injection of time. Discord/Telegram do this, TUI does not (didn't check others)

Was testing a small local model in TUI and was getting very frustrated why it couldn't just load today's damn notes! What a dumb model! How HARD IS IT TO LOAD.... oh date is gone I see

[tall prairie]

lol

How small of a model?

I saw the issue. Lgtmi, are you shipping it?

[fierce marsh]

I'm going to start contributing as a learn more here. I'm a pretty heavy dev. This software is amazing.

Nice to meet you all.

Best way to contribute is to just pull repo and make pull requests yes? Vs forking.

[jolly wolf]

You have to fork to make a pull request lol

[reef gale]

and then get the pr in before 1000

[lethal lagoon]

Briefly had a back and forth with Peter and submitted this https://github.com/moltbot/moltbot/pull/1217 before things got a bit crazy in here. I was traveling but I'm back now.

Leaving that PR closed. Based on the feedback, I've broken the work into smaller chunks: https://github.com/moltbot/moltbot/issues/3718

The idea is to add better built-in token usage diagnostics for better visibility into where token consumption happens. Curious if any clawdtributors think this would be useful.

[frank berry]

Remind me tonight and let’s discuss it

[empty fjord]

I'm probably gonna fork this tomorrow and add adjustments to this.

[tall prairie]

Spotted three PRs that look like they would be good for further review https://github.com/moltbot/moltbot/pull/3705 and https://github.com/moltbot/moltbot/pull/3677 and https://github.com/moltbot/moltbot/pull/3675.

[turbid ferry]

I'm trying to use kimi k2.5 with moltbot. please review the pr. I added kimi k2.5 model with fireworks provider

[tall prairie]

I think kimi k2.t has problems. https://x.com/badlogicgames/status/2016694973014384833?s=46

[glossy yew]

@badlogicgames via Twitter

Mario Zechner (@badlogicgames)

Is this a me issue, or is Kimi K2.5 outputting tool calls in its thinking traces on all providers at times? HF, OpenCode Zen, especially OpenRouter which seems to be completely banana cookoo.

**💬 3 ❤️ 6 👁️ 892 **

FxTwitter · Jan 29, 2026, 2:08 AM

[lone mauve]

My first PR → https://github.com/moltbot/moltbot/pull/3828

What should I do next?

tackled this one too → https://github.com/moltbot/moltbot/pull/3834

[arctic plover]

are you looking for maintainer to approve and merge them? seems like the checks aren't finished yet 😆

[turbid ferry]

Are checks mandatory to merge? My pr has been stuck in some checks for few hours... https://github.com/moltbot/moltbot/pull/3617

[timber lantern]

Hey, all. I have a PR for introducing support for Claude Code SDK (for claude code max and z.AI coding plan subscriptions). This is my first PR for the project, so I'm happy to make any necessary changes to make sure the code is acceptable to the project maintainers. I could, however, use some help regression testing a bit since introducing an entirely new AgentRuntime (instead of just using the embedded pi agent) is a reasonably complex change.
PR is https://github.com/moltbot/moltbot/pull/3823. I have yet to figure out the best way to integrate this into the Web Config UI and am very open to suggestions. Until that time, I have included sample of the JSON config at bottom of the PR description.

[lone mauve]

Yes same question as @turbid ferry are checks mandatory to merge. As it is my first time would you mind what happens a PR is created? Only after checks ran then the merge will happen?

[arctic plover]

the merge could happened only after the checks completed. I believe maintainers will review and approve it. since I am not the maintainer, I could not speak for them. However, it is just a gold standard for PR, isn't it? 😆

[turbid ferry]

i totally understand pr review and github actions are overloaded

[arctic plover]

your PR has been waiting forfor a runner for already 9 hours, OMG. but you are not alone

[grizzled osprey]

What's the current situation regarding macos ci jobs stuck for hours @storm comet ?

[clear plank]

which CI jobs and where?

[clear plank]

we have 450 PRs and new ones every like 7 minutes. we have unfortunately broken the scaling limits of a PR based process lol.

[low ivy]

ah sorry I posted a message in dev about the block mac CI jobs. I'll repeat it here
PR #2225 CI Status — macOS Runner Issue

We've re-triggered CI 3-4 times over the past 3 days. Pattern:

✅ All non-macOS checks pass (Android, Linux, Windows builds/tests)
✅ 18 checks green
⏳ 4 macOS jobs stuck in "queued" indefinitely:

• checks-macos (test)'
• macos-app (build)
• macos-app (lint)
• macos-app (test)
The PR is Android-focused so macOS checks aren't directly relevant, but they're blocking merge. Are the macOS runners disabled, rate-limited, or is there a backlog?

Run: https://github.com/moltbot/moltbot/actions/runs/21470172355

[clear plank]

its probably fucked because everything else is on fire all the time lol. high project velocity + high project attention + small maintainer team == stuff will break 😄

[low ivy]

Yeah thats what I figured, moltbot is managing and keeping it mergable so no rush 🙂

[clear plank]

for now just dont worry about stuff like that imo. most important thing is that main is mostly not broken. if you call things out like this in a PR so a maintainer can quickly scan this and go "ah, CI broken -> that explains why" -> this makes peoples lives easier -> higher chance of your stuff being seen in the chaos 🙂 thats the key thing to optimise for imo, appropriately + efficiently describing your changes to optimize for minimum attentionspan (lings cars in your PR, bad, telegraph mode description -> better)

[worldly wigeon]

https://github.com/moltbot/moltbot/issues/3319

I have opened a feature idea here for extending compatibility outside non macOS systems

[fierce harness]

https://github.com/moltbot/moltbot/pull/3672

Supports running multiple Discord bots through a single gateway. Each bot maintains its own session and memory. Bots can operate at the same time in the same channel via @mentions and direct messages are handled with independent conversation threads.

[cold cobalt]

https://github.com/moltbot/moltbot/pull/3768

allows a feature that Pete added (subagent thinking overrides) to actually work.

While I was in there I added the model override to the zod-schema for tools as well so it’d pass validation and could be used. Both were supported but passing config validation.

Will locally test it later this evening but gotta get on the box to do it since I’ve been running off stable releases up until now.

[hot ridge]

hey @jolly wolf I just wanted to share this letter I sent to Peter, still waiting on a design decision so I can submit a PR for this. I just need someone to let me know what solution we should adopt for this and I can have the PR submitted today for it:

Hi Peter,

Thanks for following up on this. As mentioned in my earlier message, the issue still persists - the link detection feature is indeed consuming messages and preventing them from reaching the AI, regardless of the domain.

I thinks that it's more of a design consideration rather than a traditional bug - it's fundamentally a matter of opinion about how the feature should behave. The core issue is that users lose the flexibility to have both behaviors: custom script processing for certain domains AND AI processing for other domains.

I've been considering a few approaches to give the configurator more choice in how it should behave:

Solution A: Add a simple boolean flag to allow the AI to continue seeing messages with links, regardless of domain filtering.

Solution B: Implement a domain filter array so the system only redirects messages matching specific domains to the custom script, otherwise sends to AI.

⭐ Solution C: Pass the original message as a second argument to the custom script, allowing the script to implement its own domain filtering logic and decide whether to process the link or forward the original message back to the AI.

I actually reached out to Shadow in the #1456350065223270433tributors channel for guidance on this design decision but didn't receive a response, which is why I haven't moved forward yet.

What are your thoughts on these approaches? Solution C seems appealing as it provides maximum flexibility to script developers while preserving the original message flow.

[jolly wolf]

on a bit of a hold today, hopefully will be able to sit down and look tonight or tomorrow

[hot ridge]

I just need letter A B or C it's that simple

I don't need you to commit to a design decision right now just curious on your opinion on which way it should go, as I made clear in that message I'm particularly fond of C however thinking bout piping a potentially complex msg through the shell to a custom script might be tricky to do as well so there's that consideration

[jolly wolf]

its not that simple bc i have no clue what this is like rn in the codebase

[hot ridge]

if you want i can provide you my design document via DM so you can give it to your AI to analyze the changed files related to this feature

like it cites all the files that were changed to support this and how

ok i did send you the first version of that file like a week ago, I just resent it via DM for the latest updated version of this file

Thinking it further we could base64 encode the original messages s the second argument to avoid any escaping/encoding issues as we pass them through he cli but i don't know. like solution A is the simplest the most "dumb" solution, solution B is the most complex, and solution C would be the most elegant but in the meantime clawdbot remains "broken"

[unkempt nexus]

How about this as a simpler solution, we could just define a new protocol where the custom script outputs FORWARD_MSG instead of having to pass the message around the receiving side can just detect that string and pull the message from context to pass through

Using a special protocol string like "FORWARD_MSG" that the system recognizes is much simpler than passing the message around. Here's why this approach is appealing:

1. Simplicity: No need to encode/decode the entire original message
2. Clean Protocol: The custom script just outputs a recognized keyword when it wants to forward
3. Context Retrieval: The system already has the original message in context to forward when it sees the keyword
4. Minimal Changes: Only requires the system to recognize the special string and have access to the original message
5. Security: Eliminates any encoding concerns since no message content is passed back through the script
   The workflow would be:

• Link detection triggers custom script with URL
• Custom script processes the link as needed
• If custom script wants to forward the original message, it outputs "FORWARD_MSG"
• The system detects this special string and forwards the original message to the AI
• Otherwise, the custom script's output goes to the AI as normal
This is definitely a cleaner approach than passing encoded messages around. It's simpler to implement and maintain while achieving the same goal of giving the custom script control over whether to consume the message or let it continue to the AI.

[summer spade]

Opened a somewhat selfish PR to add Mistral as an embedding provider: https://github.com/moltbot/moltbot/pull/3608

[hot ridge]

i switched embedding providers yesterday to a new one called moss that's not supported internally by clawdbot and it's pretty damn good, free tier allows for 3k documents

[summer spade]

That's cool but I work for Mistral so I get free unlimited API access lmao

[hot ridge]

lol oh nice congrats man

[summer spade]

But Mistral's embeddings spec is very slightly different from OpenAI's embeddings spec so it's not compatible out-of-the-box

[hot ridge]

how much data are you planning to embed?

[summer spade]

All my clawdbot's memories

[hot ridge]

have you measured the size in terms of tokens or kb/mbs?

[summer spade]

No, it's not that much now because I'm not a super heavy user yet but I've got some stuff I'm turning on that'll turn my usage way up

[modest steeple]

https://github.com/moltbot/moltbot/pull/2419

i have a PR that integrates a private, TEE-based AI provider called Maple AI. The PR was implemented based on the Venice AI implementation and works in my testing. We have users who pester me daily asking when Moltbot will have this available. 🙂

hopping in here to see what it would take to get a low-risk PR like this looked at and on the path to getting merged

[jolly wolf]

@marks might be best to build it as a third party extension using our SDK if you want it out fast

oops

@modest steeple

[modest steeple]

could you point me at a place i can read about going that route?

[jolly wolf]

https://docs.molt.bot/plugin

api.registerProvider

[stone field]

not sure who I can talk to but one of the core plugins is not published to npm

[jolly wolf]

Answered in dms 👍

[sudden sedge]

Is there a source code or plugin that would allow me to connect my phone as a node? My MoltBot says it can help but I need a source code for Apple AppStore build. I’m learning, like a young pup in this here world.

[jolly wolf]

just use the node command

[viscid jay]

@sudden sedge, docs are available at https://docs.molt.bot.

[sudden sedge]

It’s an iPhone, my bot says I can connect an Android phone easily. But right now I’m using iPhone. I use Android sometimes also just not atm

I should be more specific maybe. I’m communicating with my bot through WhatsApp on my phone. I want to connect the phone as a node so when I ask for certain tasks it can control my phone and do those things for me. If that makes sense.

[jolly wolf]

oh connect your phone

there are #1458485208721395843 that arent finished yet

[full blaze]

Can someone review? https://github.com/moltbot/moltbot/pull/3492

Trying to make it easier to visualize what account is being used for those of us with multiple anthropic/openai accounts

[turbid ferry]

My PR is pending in queue for 20+hours. Can anyone take a look? https://github.com/moltbot/moltbot/pull/3617

[jolly wolf]

you and 400 others

[summer spade]

Shoutout to a 1.5 weeks ago before molty exploded in popularity and my PR got reviewed/approved in like 2 hours lmao

[turbid ferry]

first mover always wins!

[dusky blaze]

Remember this first week(s) of January lol

[quiet shadow]

@jolly wolf hey i know the repo situation is crazy atm but you think can you guys have a review on this? https://github.com/moltbot/moltbot/pull/2923 posted couple of days ago here and many people were interested, i think it brings a very cool feature to the groups

also who do i ping to get contributor tag here? had one pr merged a month ago for linux stuff https://github.com/moltbot/moltbot/pull/541
(btw damn that one month feels like years for the whole clawdbot lore)

[ripe viper]

Hi: AWS Bedrock Inference Profiles Support
Issue: Moltbot v2026.1.29 only discovered base model IDs from AWS Bedrock (e.g., anthropic.claude-sonnet-4-5-20250929-v1:0), but not regional inference profiles (e.g., us.anthropic.claude-sonnet-4-5-20250929-v1:0). This caused "Unknown model" errors when trying to use inference profiles, which are required for some Bedrock models that only support INFERENCE_PROFILE invocation type.

Root Cause: The discoverBedrockModels() function in src/agents/bedrock-discovery.ts only called ListFoundationModelsCommand but didn't query inference profiles.

Fix: Modified bedrock-discovery.ts to:

Import ListInferenceProfilesCommand from @aws-sdk/client-bedrock

Call ListInferenceProfilesCommand after discovering base models

Map each inference profile to its base model to inherit metadata (input modalities, reasoning support, context window, etc.)

Add discovered profiles to the model catalog with proper naming

Result: Bedrock discovery now includes both base model IDs and regional inference profiles. Users can now configure models like amazon-bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0 without manual catalog edits.

Testing: Verified with AWS Bedrock Claude Sonnet 4.5 inference profile in us-east-1 region. Model discovery, configuration, and Telegram responses working correctly.

[inner sigil]

https://github.com/moltbot/moltbot/pull/4299

added some missing openai TTS voices. easy 5 line change

[mossy sky]

Would anyone contributors open to discussing this issue: https://github.com/moltbot/moltbot/issues/4305

[inner sigil]

@jolly wolf can i also get clawdtributor role? i am here in contributors list 🙂

did this PR a while back https://github.com/moltbot/moltbot/commit/333832c2e1147b9f31996849a1360fac40fb2323

[gleaming rivet]

Me either man. Your contribution is great.
A lot of people test Moltbot, but only 330 people fix and improve stuff.

This hype is crazy guys, and scary. Using Moltbot people are literally learning what agenting coding is.
And do a lot of mistakes, because gave too many tools for these agents.

Crazy, guys...

How do you feel being contributor to the Clawdbot?
Interesting how the Author does feel about it...

[gleaming rivet]

Some time ago I fixed "bash" tool that wasn't working on Windows. It wasn't return result from "cmd" or "powershell" execution.
Since this fix - it now called "exec" (Author reworked that part compeletly).

It become to be useful for me and couple of my friends, after that fix.

[lone mauve]

https://tenor.com/view/mr-bean-mrbean-bean-mr-bean-holiday-mr-bean-holiday-movie-gif-3228235746377647455

My PR are still waiting to be merged

[untold pollen]

Sorry we'll get to them, just cleaning up a few things and focusing on bug fixes 😅

[sacred swift]

Same here 🤣 it's a simple update to kimi code to support kimi 2.5 hehe

https://github.com/openclaw/openclaw/pull/3155

[full blaze]

Feature request: https://github.com/openclaw/openclaw/issues/4410

[lone mauve]

https://github.com/openclaw/openclaw/pull/4412

Why most of the recent pull requests are having failed checks like this?

[jolly fiber]

contributing.md needs an update now that we have more maintainers 🙂

[jolly wolf]

LOL

maybe a bit yeah

[jolly fiber]

how did 600 PRs not catch that

[frank berry]

I wonder if a single PR so far has been written by a human

[jolly wolf]

yes

i did one where i fixed a typo

[jolly fiber]

or you could just ask claude to "MAKE NO MISTAKES"

[jolly wolf]

slop

codex is king

[frank berry]

Kinda implies that there’s been 2 then.

[orchid echo]

Hey team! Big congrats on the rebranding to OpenClaw! 🎉 The new site looks awesome.

I just wanted to report a small broken link I found on the homepage.
The "ClawHub" card seems to link to clawhub.com instead of clawhub.ai.

Just a heads up! Keep up the great work. 🚀

[low schooner]

hey, i want to work on making it more robust on linux and also we ui/ux design of the control plane, where would be best to start?

[wise tinsel]

I want a openclaw plan mode

Great job to the team, insane pace 👍

How do you even deal with so many PRs

[dreamy quarry]

The Claw handles the PRs.

[narrow verge]

is this clawdbot community?

[wise tinsel]

THE CLAW

[spark kelp]

Hey folks just wanted to drop in and say thank you for an awesome and very useful project 🙏

I work over at Chutes.ai and I noticed the openclaw naming 🔥 so I grabbed a fresh fork and made a branch and PR for adding Chutes as a provider.

Feel free to DM me if you have any questions

[mental zealot]

Hey team, wrote support for my company Nebius Token Factory inference provider please help merge my PR … would be great to set up openclaw for a upcoming dev-rel

[spiral quiver]

Hi folks. I opened an issue to discuss adding a database tool plus a companion skill to openclaw for large scale structured data analysis (starting with MySQL or TiDB). I would love maintainer feedback on scope, placement (core vs plugin), and how to split this into small reviewable PRs. https://github.com/openclaw/openclaw/issues/4605

Background PR: https://github.com/openclaw/openclaw/pull/2704

[knotty zinc]

Any one down to collab on a linux companion app to have full clawdbot support including canvas, screen record, system.run, camera, and voice? Or is that already on the roadmap?

[clear plank]

interesting idea. my initial reaction: we will most likely end up on a lean core model, so i would design with that in mind - consider making your own OSS cli and then exposing it as a skill + CLI. exact extension model for this kind of thing is not ready yet, we are aligning internally on how to do it (its hard, since you are basically coupling RCE with prompt injection at scale, as a service, but pretending its a package manager). imo most bots do not want a DB, so it wouldn't belong in core but its a nice opt-in feature to have. thats my reaction after spending 1m reading the issue 🙂 but you are thinking the right way.

[low ivy]

As name changes are the cool things, ziggy_moltbot has become "Ziggy StarClaw and the Lobsters from Mars". Its a independent Zig implementation of the client (operator so far) protocol for Windows, Linux, WASM and Android. Its very early and ugly but all platforms can connect to OpenClaw gateway and chat. More features and a UI that doesn't actively hurt you soon. https://github.com/DeanoC/ZiggyStarClaw

[tall cairn]

https://github.com/openclaw/openclaw.ai/pull/31 There is still a mismatch in the installation description on the website. Its still mentioning "clawdbot" instead of "openclaw".

[simple token]

Is anyone aware of work/plans to support LM Studio's new v1 rest api? (please correct me if wrong channel)

[quartz bolt]

dont go to po**hub if it is not worked!!! lamo

[hasty fiber]

Surely "clawtributors" instead of "clawributors" 😅

[quartz bolt]

its clitributors 😄

[noble python]

Hi all trying to add nova as a model provider. https://github.com/openclaw/openclaw/pull/3512/changes
was wondering what I should do next

[misty turret]

Worked fine

[gusty shore]

Found a bit of an issue with the current commit in main - some approvals stopped working/showing and auto-denied. Anyone else seen that?
https://github.com/openclaw/openclaw/issues/4725

[stone field]

looks like the Tlon plugin is still not published to npm, how can we get that published? its currently broken in the current flow since it cant find the package

[toxic ivy]

I found it quite difficult to change API keys or register new ones once the setup was already finished. After some trial and error, I managed to get it working, but what do you think about adding a feature that allows us to add or reset API settings directly through the CLI or a dashboard?

[summer spade]

It's part of the standard onboarding flow now afaik

[stone field]

what do you mean exactly

[misty turret]

Hola amigos! for those who work with bedrock in aws using inference profiles i just push this https://github.com/openclaw/openclaw/pull/4757

[cold cobalt]

finally was able to locally test and update the fork/pr to match both renames we went through

passes a daemon/gateway restart now without issues with the config. checks running now.

[turbid ferry]

bump up pr again for adding fireworks ai provider. https://github.com/openclaw/openclaw/pull/3617. the fastest provider for kimi 2.5 atm. I'm looking forward to using kimi 2.5 with openclaw. credit: https://artificialanalysis.ai/models/kimi-k2-5/providers

[solemn lion]

any help here would be appreciated

[summer spade]

Oh huh. I'd do a hackable install then https://docs.openclaw.ai/help/faq#whats-the-difference-between-the-hackable-git-install-and-npm-install

[frank berry]

Wondering if I could get a maintainer to look at https://github.com/openclaw/openclaw/pull/4887

I've got other PRs open, and this is a brand new one, but there's a bit of a security edge to it with context bleeding over between sessions. I think it'd be worth getting toward the top of the review/merge queue.

[steel zephyr]

Is it possible to have a PR merged in this repo? hahaha

Almos 800 PRs already

It looks good and bad at the same time

[valid pumice]

Hey I’m open for collaboration or just to help in Genral

[final forge]

Posted an RFC for exec command security (catches dangerous patterns before they run). Feedback welcome! https://github.com/openclaw/openclaw/discussions/4981
I have a PR ready to go.

[valid pumice]

Sounds like your guys need reviewers ?

[solemn lion]

thanks that doesn't help users trying to install from the setup screen. we just need it officially published because they are running into errors at setup

[stark bane]

I'm trying to install openclaw using nix-openclaw on my mac, failing miserably. But I've fixed an issue I had on the code (a simple broken path, fixing image related error)
https://github.com/COLAMAroro/nix-openclaw/commit/4c43bb5cb2b46c75207a41ca4390ba0801da8cc9.patch

(reporting the issue here because no PR policy)

[lime grove]

Ok if I post a PR for session lifecycle events? It looks like you were planning on adding it anyhow.

[steel zephyr]

Can I has code review? https://github.com/openclaw/openclaw/pull/4125

[sacred swift]

Bumping this too :3 - https://github.com/openclaw/openclaw/pull/3155

[cold cobalt]

you have some CI failures FYI - probably should be "text", "image" rather than "text, image"

[sacred swift]

oh i see :3

thanks

[sacred swift]

i think most of them are passing now

[south socket]

@calebwinston_13934 warned

Reason: No advertising other servers
Duration: Permanent

[chrome thunder]

Sorry! It's a FOSS project building on openclaw but I get why it is a rule so will abide

[proud escarp]

holy crap 800 pull requests lol im praying for you peter & core team lol

[hot ridge]

hey @jolly wolf just follwing up. I never got a response on how to handle the tools.links change I proposed a few different solutions, Peter said he was too busy to think about it, you said you needed like a day or so. Lmk if you find it. If i were to implement a solution for myself it would be **solution D **

I'll paste solution D from the bottom of the document here:

Alternative Approach (Solution D)

• Concept: Define a new protocol where the custom script outputs a special keyword like "FORWARD_MSG" to indicate that the original message should be forwarded to the AI.
• Implementation: Instead of passing the original message as an argument, the system would recognize the special keyword from the script's output and automatically retrieve the original message from context to forward it to the AI.
• Advantages:
  • Simpler implementation without encoding concerns
  • Cleaner separation between link processing and message forwarding
  • Eliminates potential security issues with passing complex messages through CLI
  • Minimal changes needed to the system
• Workflow:
  • Link detection triggers custom script with URL
  • Custom script processes link as needed
  • If script outputs "FORWARD_MSG", system forwards original message to AI
  • Otherwise, script's output goes to AI as normal

[marsh barn]

dont forget (dont forget) to collect my noteboos

[mental zealot]

Guys I think branches are not being merged in order and creating conflicts on PRs which had no conflicts before and lot of fixes to be done due to this, had to close one more PR without being merged 😭😭😭

[elfin orbit]

Need reviews please: https://github.com/openclaw/openclaw/pull/3616

[sonic hinge]

Hey, i am working on this version: https://x.com/srisanth2004/status/2017550308574990587

Looking for help creating lite weight version of OpenClaw...

[rigid trellis]

new playground for AIs

clawnet.org

they can find jobs and make money, earn reputation

[toxic ivy]

I am interested in becoming a contributor.
Could you point me toward the right person to discuss this with?

just add issue and wait is right?

[split nexus]

In general, add issue, fix, PR, post here for review, when merged then request clawtributor badge/status.

(Unless the process has changed in the last week)

[split nexus]

☝️

[clear plank]

we have like 1k issues and 1k PRs and a new one maybe every 5 minutes. this is basically impossible for maintainers to go through. dont expect much to get merged any time soon. the only way we are going to be able to handle this is using AI tooling to "mow the lawn" of PRs, and there will be some good PRs caught by the AI too. you'll have to adjust expectations accordingly - we still value all the contributions, its just, well, impossible for humans to manage this many

this is quite possibly the repo with the fastest PR/issue creation rate on the planet right now i think. maybe claude code has more. idk

defintiely top 10.

[untold pollen]

Hi, can someone running OpenClaw on Telegram check if they see this toggle in @BotFather?

You can easily enable topics in private chats by toggling on Threaded Mode via @BotFather.
https://core.telegram.org/bots#natively-integrate-ai-chatbots

I have like 8 bots and I don't see it on any of them

Ah nvm, you have to use the new BotFather app for this

[timber lantern]

Hi, everyone. I opened a discussion w/a potential design for abstracting an AgentRuntime so we can support both Pi Agent, Claude Agent SDK and other potential agentic frameworks down the road. I had previously finished an implementation but it was messy and needs to be factored out into a few PRs. It sounds like the right first step is to share an RFC/Proposal which I just published. Feedback would be welcome. I'd really love to get support for CCSDK for our agents since I hesisitate on using the CLI as I've heard rumors of people getting banhammered by Anthropic.

(also if there's a better spot for this to get posted, please lmk 🙂 )

[untold pollen]

God we're probably going to hit 1K open PRs soon 😵‍💫

[hidden fern]

I made a few changes. Computer generated token and auto connect gateway on startup (that was a point of confusion for myself and a lot of users) and broken logic for the ui WhatsApp QR code connection. Both for the connection process using the QR code AND maintaining the connection with the key stored on host. How do I push it as a PR?

[hidden fern]

You sent that as I was writing my text 🤣🫣

[clear plank]

give it like 20 mins

also FYI everyone we will soon use AI to manage our open PRs. if this hits your open PR, dont be offended. we have one PR every 2 mins and its impossible to have human eyes on them all. also while lots of PRs are decent, lots of PRs (probably not yours) are lazy and bad and just AI slop. so we gotta keep a lid on it to keep maintainers sane. basically, congrats, we broke github. we are the kim kardashian of open source.

[hidden fern]

Damn, I guess that’s a good problem to have! That means a lot of people want to see this succeed! Excited for the days to come!!! 😎🦀

[terse lance]

https://substack.com/@samillingworth/note/c-207832798?r=47fdmh

What do AI-only social networks look like? 🦞

In this new post from Slow AI, I explore what Moltbook, a new AI agent-only social media platform, is and what it means for humans.

TL;DR the reflection in the mirror ain’t pretty.

[mental zealot]

Openclaws team made the same PR the 3rd time to add Nebius inference provider … all tests passed with no conflicts at this time 😂… somebody please help ..
https://github.com/openclaw/openclaw/pull/5508

[lone mauve]

I have 9 PRs waiting in the queue for more than 2 days

[split nexus]

How can we help point AI tooling at this?

[thorny lark]

I'm sure they'll figure it out, bud. It's like we're managing our emails at the moment through clawdbot, they just need to manage PR's and issues through clawdbot.
Give it little time

[modest steeple]

still hoping to get some eyes on this PR. it's very similar to the Venice AI PR from an implementation standpoint. provides end-to-end encrypted AI provider Maple for extra security. i've been running it all week and it's working well: https://github.com/openclaw/openclaw/pull/2419/

[wintry stirrup]

opened a PR for the Usage tab (#5520), wires up the existing usage.cost/usage.status RPCs that didn't have a frontend yet.

[topaz coral]

Any discussion of multi-user agents or similar setups yet? Thinking of taking a stab, I know the backlog is huge