#jules-general・🦑
1 messages · Page 8 of 1
Hello everyone, how much data can Jules store in its memory?
Im building jules react native app for ios !
how to add a setup script for Jules? I don't see this anywhere:
never mind - found it. what a way to hide that
I just wanna give a big thanks to the Jules team. I've literally been gone a week and a half (since the Pirate event ARRR! 🏴☠️ ) and so many features have been added. It's really good to see that the developers are actively taking feedback.
I have a list of github issues that I want Jules to complete in a sequence, how can I achieve this? The simplest way I can think of is manually managing this by adding the jules label. However, is this something I can have Jules take care of?
context window?
i
Hi guys
Unfortunately, that doesn't seem to work.
All it does is create a modal on the web interface that requires me to click the "Publish" button by hand 🙁
The API is incredibly useful, and I now have an N8N task that creates a new Jules session from every ticket in a particular Trello list.
However, there is one big roadblock - and that's that I would like Jules to automatically push the branch to GitHub (so that the CI can be triggered.)
Unfortunately, even for tasks that are created by the API, Jules still appears to require a user to go to Jules.google and click the "Publish" button.
I don't know which of the following are true:
- This is intended behaviour for the API 🙁
- This was intended behaviour for the UI, and there is an API call / parameter to do this but I've missed it.
- This isn't in the API, but will be in a future version 🙂
At the moment, this means that the programmer whose credentials are used in the N8N task is the only one who has visibility of the change, because only they can push the new stuff to GitHub.
doing the same
hello guys, can some one help with docker setup for jules ?
Hi there!, could you plz share more details about your Jules setup or the specific issue you're facing with Docker?
Hello! To clarify: I run the project locally with Docker. When Jules works on it and applies run-related changes, it can’t verify whether the service actually starts. As a result, I have to repeatedly fix issues that Jules could resolve automatically if Docker were available. The docs suggest Docker support exists, but I haven’t been able to configure it successfully yet.
i actually found a way. click the link it gives you authenticate locally, copy the redirect link, then open another session of your vps while the first one is still waiting then use curl "xx"
Oh that’s an interesting idea. Let me try
I tried port forwarding
Actually now that you authd can you tell me where it put the auth credentials? Is it in ~/.jules?
I noticed that whenever I start a new session, Jules runs my setup script. Isn't that the point of the environment snapshotting after you "Run and snapshot" in the environment configuration page, that you don't have to run that and can just clone the snapshotted image?
Give the template pls
The snapshot should save your setup so you don’t need to rerun it. If it still runs, the snapshot was probably taken too early or not being used. Try rerunning setup, then “Run and snapshot” after it finishes.
Yeah it worked successfully and said saved snapshot with the green Success message, but every time I start a session it runs the setup script again
Also , is there any way to set up a team and invite other teammates to your Jules repo setup? I feel like there has to be and I'm missing it, but have been poking around the UI and documentation and haven't found anything. I just want them to share the repo I set up and created the configuration setup script for
What’s the best way to share an n8n template? (Something I’ve never done before.)
I could write it up as an article and share a link, but I don’t want to spam…
… but there may be a more n8n way to do that?
Just download the workflows and send it !
Until Jules is able to GPG sign commits, I'd be open to allowing it to bypass the branch restriction. Unfortunately, just adding the bot to the bypass list doesn't seem to be helping. Anyone successfully been able to do this, or any help on what might be needed to make this possible?
I'm currently working around this by utilizing Jules in a separate fork then bringing over the commits via my GPG key. I don't think they would implement GPG or signing in general right now since it's a bit of a grey area in terms of implementation. Potentially an enterprise edition of Jules?
Would be cool to see in this month's Workspace update
An option, but not one that I think is going to be easy to scale to other engineers. I don't need Jules to sign commits as me, just that they are signed and "valid" as far as GitHub is concerned. In the past I've has success getting GH to sign commits using its GPG key using the API, but that requires a bispoke implementation that doesn't integrate with the normal git cli/tooling that is out there.
Is any on Gemini pro feel like coding is not doing so great job with context and design and layout feel like Jules have been not acting like it self. Think maybe guardrails is making more stupid anyone else notice that is get last or functions worst?
doing fine for me with php, html/css/js, and python.. its pwning
maybe your code is super complex, but I do mini edits like 50-100 line changes and commit each time and its awesome.
if u trying to do full refactors or using like 1m context then idk
might be context/memory rot
i just wish it was a bit faster, maybe gemini 3.0 soon 
this might sound silly but jules will occasionally make silly mistakes, it's important to let it finish the task and for the review agent to do its job. for example in that pic it messed up a job and the review agent noticed before the commit and PR. Jules went back for an hour or two autonomously fixed everything and then pushed the full, correct implementation
Thanks for the update! We strongly recommend you always allow Jules to complete its self-correction cycle.
for me it doesn't want to stay logged in. as soon as I authed, even with this method, and then run jules, I get:
🦑 Welcome to Jules CLI! v0.1.30
What would you like to build today?
Failed to list tasks: Trying to make a GET request without a
valid client (did you forget to login?)
Tip: you can use "/login" in the TUI or "jules login" in the
command line to login
> Search sessions or type / to use commands
The task table is empty. Create a new task using the /new command!
Working in: ~ unknown/unknown (unknown)
login in through the TUI at least lets me use jules until I close it, then it's again no longer authenticated 😐
How? Getting this on almost every task
Getting this a lot with html webpage just one landing page.
Getting this alot, Simple landing page in HTML. Any suggestions?
It looks like you're providing HTML files for Jules and Jules can't execute any commands because there aren't any. In a modern webdev environment, it would run linting and type checks then execute something like 'npm run dev' to start the web server emulation and test that the dependencies and code logic was functional.
Why don't you try throwing a a more difficult task at Jules like , " refactor this website in to a modern product showcase page that uses tailwind and next.js stack. "
Once pushed to GitHub you can then deploy this to Google Cloud or any other web hosting platform
You could also use one of the web authoring tools like Stitch or Firebase to create the front-end design yourself, then get Jules to pull the information from your old landing page and put it in for you
In short, you've not provided Jules the entire code base or instructions on how to set up and run Apache/MySQL to host the WordPress website? Once it has completed editing the html it checks for syntax and coding errors but that's all it can accomplish without a backend of some kind
It's fail anyways it want me too restart the session
this is just a set of rules for a coding agent.
you've just given Jules a singular .html file... you can't just "run" or "execute" it like a video game to make it do tricks - it's not built for that.
think of a backend as the "kitchen" in a restaurant: the HTML page is the menu (what customers see), but the kitchen (backend) prepares the food (data) and serves it up.
your .html file needs a backend stack like node.js/express, python/flask or php
Yeah I give refence and data and where it can find the info
Look at this page kihlstroms.se . @sharp flame thinks it great job sometimes it can do amazing work and write good html code and php and scrap sites for context and almost be done then it like collapse because of workload. Or delete the whole project and do webpage that takes like 1 hour of code or less very Simple like it lost all it knowledge. I had trick using memory before it was available
Going rebuild website it got real update 2022
So Design improvement is need and functions it is based on WordPress @sharp flame if you have suggestion and thought it would be much appreciated always fun hear what other think.
wordpress usually relies on a database like MySQL for images and user auth
Jules can only edit so much of the HTML because you have not given instructions on the LAMP (backend) stack . any CRUD changes to the frontend will break once you deploy to your webhost .
I gave it links and photos and content and context usually and refinements but maybe that wrong way use it. It look good costume php webpage for WordPress when costume webpage with code. But probably not right way optimal. Just started like fun project because webpages is in so bad state and theme is from 2022. So Design improvement is need. Looks WordPress 6.9 will be upgrade for using it ai but it comes out in December 20 oktober
Beta preview
jules is for big development tasks. for your case i think maybe you need to use Google AI Studio with the URL Context enabled
Have some big development plans but don't wanna throw to much at it if it not ready
Tried google ai Studio but it has to much hallucinations and collect wrong faulty data because it is in swedish language for most part. The support not so great for sweden
sounds interesting
I disagree with this. Jules is quite literally meant for targeted well defined E2E tasks, I think in larger scale tasks is where it stalls significantly due to it's architecture. Unlike a HIIT implementation it doesn't have full context of your codebase, it's new memory function does help but it's still very limited in it's ability and I don't think it should be relied upon to do large development tasks.
As @sharp flame mentioned, and as an ex-Wordpress developer it's difficult to get Jules to integrate well with your intended use case. The LAMP stack is very different to the SPA architecture or other environments Jules expects to work on [https://jules.google/docs/environment/]
You're better off in my opinion using Wordpress in it's headless CMS state with a React front-end which is where Jules would excel. But you'd lose the ease-of-use that WP is known for.
How do you think best way prompt it or use reference
trying to update my snapshot and it keeps saying the working tree is dirty
Hey @visual gyro Jules can’t run commands since you don’t have backend or setup provided. Try assigning a full task like refactoring it into a Tailwind + Next.js project, then deploy via GitHub. Without a full codebase or server setup, it can only edit HTML and check syntax.
Hello @orchid jasper Looks like there are uncommitted changes in your environment. Try committing the changes or discard local edits. Once the working tree is clean, the snapshot should update fine.
I tried that also but it freeze the tasked. I even gave it html to edit and check syntax and use html as guide as working on task and project. Feeling like lost alot of power when it went our of beta version it could handle bigger project before it feels like it became lazy? If you look at all my previous task you can see big difference
is there some plan to be able to get custom environment, so we can have custom tool to run inside Jules like dart/flutter without installing them every time?
Also would be great if we could use gemini extension in some task, is there a plan for that as well?
The freezing could be from the new snapshot or environment limits. Try resetting the environment or creating a fresh one and re-uploading your project files.
How can tell a limit on environment what is the issue with the new Snapshot? Folder 📁 📂 and files coming I would like name project and different task renaming them
im in the same position. link i get while in TUI is partially clickable so its broken. so i login with jules login instead of /login but i cant use it after
Once I've created a Github branch for a task, and then asked for more changes, how do I make Jules add those further changes as a new commit to the existing branch?
Please check the environment’s resource panel—freezes usually due CPU or memory limits. The new snapshot may have missing dependencies, so try rebuilding it. Name your main project folder and add subfolders per task to keep things organized.
"It seems like you've been away for a bit. Are you still there? I'm ready to continue with the next step whenever you are."
this seems to be occurring quite a lot lately, any tips for getting it to stop? Jules used to be very good at autonomously running until task was complete or the review was done and awaiting further instructions.
Wondering when it will be available in Morocco!
Thanks for reporting. Will forward this issue to our internal team. Meanwhile, we recommend you to break large tasks into smaller, bounded steps to avoid resource bottlenecks.
Please update if it figure it out
Sorry but during beta I pushed considerably larger and more complex tasks with several phases through to Jules without problems. When was the decision made to turn this async agent in to something that could only handle piece-meal, step-by-step small tasks? Who made this decision specifically?
What's the difference between tasks and sessions?
my jules is keep geting stuck on the git push phase
the chat somehow doesnt render the approval button and then the whole session just breaks and glitches
Hey @sharp flame To be clear, nothing changed in the product's core capability/no one made any changes. Jules is still designed for complex autonomy, that was just a temporary workaround suggested to unblock. Please note that our team is actively working to resolve these issues, you can open a thread in #1374125660859142254 channel
On my end or in cloud? Have the same issue as @sharp flame think Bottleneck is happening alot more now even small one page html with only 3000 lines of code
Think it lucked behind 🔒 ultra paywall Jules feel less accurate prefer 4.6 model and cloud 4.5 feel more accurate prefer it for the moment. Jules feel very inconsistent right now like it lost it ways @leaden pawn
Please report this along with task details in #1374125660859142254 channel
PRs seem to be completely broken right now for me
Failed to create pull request: GitHub API validation error: {"message":"Validation Failed","errors":[{"resource":"PullRequest","field":"base","code":"missing_field"}],"documentation_url":"https://docs.github.com/rest/pulls/pulls#create-a-pull-request","status":"422"}
Has anybody tried connect jules to linear yet?
Hello @swift karma A session is like your project room and a task is a single job that you ask Jules to do inside that
ICYMI we have an AMA tomorrow: https://discord.gg/googlelabs?event=1422327008335102113
Ready for all your questions 😀 . Also if there's a specific demo of our features you wanted us to show please let me know and we'll that task prepped for tomorrow.
would love to see the new api work with the linear api!
WordPress site website and Merchant and Google Analytics connector and different api plugs and connectors and seo analysis or something that can help small start ups
How can create svg using jules please explain
how to setup the environment for a repo where some python libraries have to be installed?
does Jules do it by reading the readme.md ?
Just tell jules to make one
I dont see that much memory if I try to display all the list. Is this a good sign ? Is it possible there is hidden memory or Jules lie to me ?
In which countries is Jules available? I can't access it from Ukraine, even though I bought a Google subscription.
I use uv and put my requirements in pyproject.toml jules does a good job of installing them
I bet it would also do a good job if the libraries were in requirements.txt
has anyone had real success with jules yet? it reads my repo and fails at random places
yea I did it handled my lua project very well
was that a greenfield project or existing repo?
my prompt was "find ways to reduce duplicated code"
It's my time to try Jules.
Yes please go ahead and let us know if you face any challenges
hi, i have switched to Google One AI Pro subscription, but Jules still shows basic plan, how to update its plan?
it should show wen you sing in with your google acc tie to the sub
Hello Jules friends
I missed it where can I see recap?
Promt please
@elfin kite can can like video recap or from event stage videos in like a #1238621511401734254 or link special deals or guide channel ? Maybe youtube channel for labs??
Do you mind if I DM you. (Let me check tho if I still have the workspace. I cleared few things from Jules to reduce clutter)
Is it a public repo? Sure do so. What was application or refactor or what was it can not remember now.
Anyone got some tools or something using api in smart way for Jules?
Question.
In a session does jukes need to publish the PR in order to keep working?
it shouldn't, but it's been acting a little funny the last day or so. if it is giving you the publish branch/PR box you should be able to just send another message with whatever you're wanting it to do next, or any feedback like "you missed xyz feature"
Hey @deep parcel No, Jules executes the task and prepares the PR, but it remains available for further work/ new requests while the PR is awaiting your manual review and publication
Hi. I have had issues with jules for a few days now. It takes many hours to do basic tasks that he used to do in minutes before. I am talking 5 to 7 hours to finish a tiny task. It looks like pre-commit steps are the culprit. Did you notice something similar?
Ah cool, if only small changes I just review in the website ui
You using the same session?
I find if you hammer the same session it keeps it all i context and slows it
You got like 100 a day I think, mix it up
It's a small task. Don't I have to keep one task in one session?
No, but if you session has lots of other small tasks in
No it's one task only but with a huge amount of bug fixes and back and forth on that small feature. Was quite inefficient to be honest. And that seems to be the case for me for the last few days. I am starting a new session every time but it takes many hours every time. Last month it would have taken minutes at most.
For instance today, I accepted the plan at 09:39. It's still busy at 16:57 and most of it is the pre-commit steps. And it's not just this one, seems to be the case for every new task these last few days.
Hello if you're encountering errors and /or bugs or would like to leave a general feedback about Jules please don't hesitate in utilising the #1374125660859142254 channel. It helps to keep the team organised and it’s also where all the technical issues are logged. The problem you’re facing may have already been posted there so please also check the channel for updates.
You can also stay up to date on Jules updates via their X(Formally Twitter) social media page at x.com/julesagent.
Till then we ask for your patience as the team works through the bugs and feedback requests
Thank you !!
Hello @wintry hare Please post more details in #1374125660859142254 channel, we will investigate further.
How do you do that WordPress?
Nope you can see all the files in the jules UI. And the diffs with the patch style markings.
Hey i was wondering if you were able to implement a progress bar while jules is completing a task . More specifically to the left of the publish button above the code editor. I feel like this simple add would be pyscologically satisfying for me and im sure others as well.
but other than that I really love what you have built. I think it is absolutely amazing and has increased my productivity 10 fold. Thank you <@&1173646484311384114>
With the failure rate of agentic coding I don't think that's viable. It would go forward then back to replan then forward and back haha. Would drive me a little insane imo 😭
what agentic coder is a failiure? Or are you just not using it right?
Because I know you're not talking about Jules. You cant be
please add inline diff view 🙂
my task stopped working didnt even use much promts
session/1417547498278013185
will i be able to reuse the chat?
It'll take it's sweet time, 50/50 chances
No it's just that logically I don't think it's sound to realistically track the progress of a task defined by Jules. It's plan in itself when tasked does fail in certain scenarios but Jules is capable of altering it's path. Doing so would change the progress back to whichever point in it's plan it diverged from. It's not a user error but just part of development. Expecting any model to realistically single shot operations isn't viable at the moment. The tech just isn't there right now
Jules is a great tool but it does make mistakes as do all AI tools. The whole premise of models is built upon estimation so failure is bound to happen.
I wonder which Gemini-Version, especiall if it is pro or flash, is behind Jules. Sometimes it makes mistakes which are so stupid, I would assume Flash, sometimes it is very very smart and I would assume Pro.
Can you clarify that, <@&1173646484311384114> ?
So I’ve been trying out Jules today. It desyncs because of safety filters when trying to run a test to verify hotel adress’.
I stopped using Gemini quickly because of the safety issues (too much!), VEO3 and nano-banana.
You create good products and destroy them with censorship!!
Unusable in my case because of adress’…
maybe i'm just not working on complex enough tasks, most of mine are 1 or 2 iterations I work with Jules on improvements and i use the interactive plan to make sure everything is aligned. Can you give me an idea of what projects maybe he lacks in or goes off track? My only real issues with Jules have been environment bugs but they have been taken care of really fast I havent had a single problem for a couple weeks but like i said im probably doing micro projects compared to most people
Is there a way I can contribute to Jules for free? @leaden pawn
i have google ai pro but i only get 15 daily tasks on jules
instead of 100
Is this because of advanced protection being on?
Clear your cache or try refreshing the page, or log out and log back in.
maybe idk
do we have an alignment problem?
Still doesn't work
Do you know if jules basic plan uses a weaker model?
It doesn't
so u are not pro
I have three ai pro accounts, two works fine
And my main one doesn't
I do have pro
Well I'll wait for a google staff to see the messages here then
Not sure if the Google one support can help with labs requests
maybe
Because there’s an issue with your account it’s not recognized as a pro account.
idk maybe
@lapis fiber

Hello if you're encountering errors and /or bugs or would like to leave a general feedback about Jules please don't hesitate in utilising the #1374125660859142254 channel. It helps to keep the team organised and it’s also where all the technical issues are logged. The problem you’re facing may have already been posted there so please also check the channel for updates.
You can also stay up to date on Jules updates via their X(Formally Twitter) social media page at x.com/julesagent.
Till then we ask for your patience as the team works through the bugs and feedback requests
Thank you !!
why cant it ask questions correctly anymore its saying it sent the questions but dosent send anything, and when you send a message the questions pop up for a second and dissapear??
@prisma vine
Based on discussions with many people in the tech sector, Jules AI is causing significant issues for us Workspace users. We have a very small limit on tasks (15 daily) and only three concurrent tasks. If we use one task per job, we deplete our limit in less than an hour, which makes it impossible for a professional to do anything meaningful with Jules.
As a result, most people add around 10 to 15 jobs per task because they have no other choice. However, this approach causes Jules to malfunction in many ways. I've tested it, and Jules is clearly struggling to handle multiple jobs per task.
The problem is that you have lowered the usage limits so drastically for Workspace Pro Business users that you are indirectly compromising your own product by forcing users to consolidate jobs.
Since Workspace users are already paying a significant fee for Google services, paying again for an increased number of tasks makes no sense. Most users are either leaving and switching completely to competitors or staying with Google but experiencing many bugs with the coding agent.
This feedback is important because all the workspace users I have been working with really cannot build anything functional with Jules, it breaks their code all the time because of the large number of task they write per single job due to the limit constraint.
Also
This is the point where most of us are stuck, and I genuinely cannot understand Google's logic. I was initially on Google One, and then I upgraded to Workspace because it was promoted as including Google Pro features for business users. Now that I've spent time setting up my Workspace, I realize that our usage limits for Jules are completely different from those of Google One users.
This makes absolutely no sense, we upgrade to a business-level account, yet we receive less usage capacity than a personal Google One account. This is extremely confusing and frustrating coming from Google.
Hi, anyone got an issue with daily tasks quota that doesn't reset after 24 hours (AI pro plan) ?
As outlined in the Jules changelog changes will be arriving for Workspace users at the end of the month.
Good afternoon everyone! I’m excited to be here.
Yes!!!! Super cool
Based on discussions with many people in the tech sector, Jules AI is causing significant issues for
Could you please show me where you see this news as I do not see anything here: https://jules.google/docs/changelog/ thanks!
Not even joking they removed it from the changelog. I can't send the link here but you can check on the Wayback machine
So very odd
It was under the entry for Jules CLI
Suuuuuuuuuuuuuuuuuuuuuuuper sketch that they're modifying release details like that without notification.
As a Workspace admin im also concerned 😂 we literally just settled our annual bill last week lol
It is scary to have such significant news amended. Now check this: it was received by email in August. When I saw this, I contacted Google Support and asked if the Workspace Business plan would include the PRO version of Google AI. They confirmed that it would, which is correct, as I see it in my account (PRO). However, after joining Workspace, I found out that the PRO version from Workspace is not considered the same as the PRO version in Google One. This is very poor advertising because once you set up Workspace and move all your data over, it is not an easy process.....
🤷♂️ Google be doing Google stuff. I guess it's the way they operate. But in terms of retroactively changing details like that it's just bad business practice. Puts a bad taste in peoples mouth.
I agree
I added to your thread but hopefully they provide a reason as to why it was removed
thank you so much Tim, very much appreciated
Hi , trying out Jules TUI on Windows 11 and I am getting "Warning: git is not installed on this machine, please install it first." when I run Jules. I have git on my machine though. Any advice? Thanks!
Or is this a bug that I should post in #1374125660859142254 ?
Any update on the task manager? Any news on automode and any news on connectors for Jules more then api function's.
Hello @near jay Looks like Jules TUI can’t find Git because it’s not in your system PATH. Please make sure C:\Program Files\Git\cmd is in your environment variables, then restart the terminal. If using WSL, install Git inside it too.
Hey @grand aspen We really appreciate you asking this! Yes, you can absolutely contribute to Jules by exploring it more and providing detailed usage feedback, thank you
Hey everyone on the Jules team, considering that Jules is associated with the Gemini membership, I wanted to ask if Gemini AI Plus subscription users will have access to the plan. I only see options for AI Pro and AI Ultra plans, I don't see anything for AI Plus.
Hi! There is a way to create an auto-PR with the API. Here is an example:
curl 'https://jules.googleapis.com/v1alpha/sessions'
-X POST
-H "Content-Type: application/json"
-H 'X-Goog-Api-Key: YOUR_API_KEY'
-d '{
"prompt": "Create a boba app!",
"sourceContext": {
"source": "sources/github/bobalover/boba",
"githubRepoContext": {
"startingBranch": "main"
}
},
"automationMode": "AUTO_CREATE_PR",
"title": "Boba App"
}'
see docs under https://developers.google.com/jules/api/reference/rest
Hello @soft yacht access to Jules has been bundled only with the Google AI Pro and AI Ultra subscriptions
Is there difference in Jules if I use pro or the free version? Talking about quality of code not the limits
Any particular reason for excluding Jules only? I see that the other products do apply to the Bundle: Flow, Gemini, whish, veo3... I understand that the Plus plan is intended to balance the value a bit in emerging markets, but it's still 2TB of storage. I would like to suggest that you take the Plus into consideration, it is precisely for markets with difficult access, it does not have to have the same capacity as the Pro, but it could be something in between.
We understand your valuable request and will pass this on to our internal team
Nope! All uses Gemini2.5pro and has the same features. Only difference is 15sessions vs 100 a day.
But it's a 24hr rolling window and 15 is a real lot of work done
I really appreciate your time and willingness, thank you.
Thanks!!! At this time 15 is good enough for me 🙌
I'm having an annoying problem where Jules appears to be working on old branches. It's referencing old functions that we removed earlier today that are no longer on the main branch. Has anyone else experienced this and know how to fix it? Wish I could actually see what commands its running to start up, clone the repo, and check out its own branch. Maybe I'm missing something fundamental here that I should have in my setup script?
yes i noticed that it works on the cached vm and can only be updated when you click the button again as far as i know!
Do you mean the "Run and snapshot" button? If so that would be brutal
yes 🤣 i am running tests now to tell him to update from the main branch the first one failed now he saied:
Update branch from main.
As per the instructions, I will fetch and merge the latest from the main branch to ensure my workspace is up-to-date before starting implementation.
But its not dooing anything ATM, ill PM you if i find a solution
Ran: git fetch origin main git merge origin/main
Command timed out.
Yeah I asked it to merge from main once and the resulting PR had hundreds of files changed (the actual change was one line) so that tracks.. It must have been every change between when I last clicked "Run and snapshot" and when I started my session 😅
i managed when i did not run it trough gh issues.
but sending it directly with an up to date branch.
probllem: click the dam button to publish
Ah interesting I was wondering if this could be the source of a lot of the errors I'm seeing (I haven't successfully gotten it to do anything). I wonder if the new API integration would be better there, will try that out
im sure they bring out more features soon. the cli app has delete sessions as well not available in the api.
Give me an hour or so ill publish jules-sonsteroids-cli
rust app that uses the api with added gh features and works greate with vscode (only tested claude 4.5 so far)
Hi @leaden pawn , thanks for the advice. However , git in both my user and in system PATH and also I can open cmd and type 'git' without any issues.
https://github.com/kiwina/gules alpha , have not tested much the mcp i use it mainly from the binary directly.
im ran into a problem, there are no submit buttom after jules work, how to solve this
You just have to tell Jules to use its submit tool and try to get it to use it which will let your publish it
@fair void great 👍🏼 thanks. Think we need one of lm notebook guide and easy way suggestions depending of knowledge. No coder, set up , basic knowledge, way different google api or services
Was it roadmap or know issues with fix in the works
Does anyone know if environment variables are accessible by the sessions as well? The wording on the environment configuration page makes it sound like they should be, but I also see a) the Run and Snapshot flow always ends with an unset command for the env vars and b) my session is trying to run the setup script and failing with a message that sounds like its because it doesn't have access to my env var
Hmm I actually conversed with the session and it appears it does have access to my env var. But something is still operating differently. The command that is failing in sessions but succeeding in the setup script uses sudo, maybe sessions have different setups for the root user?
What has changed?
You give Jules a clear task, and it still modifies other files and functions.
I explicitly say not to touch anything else, yet it does. It’s become unreliable.
All day doing that - Fix one thing, and it breaks others.
Hey @primal lake Yeah, sessions do get your environment variables — the issue is with sudo. By default, sudo doesn’t pass user env vars, so they won’t show up unless you use sudo -E or adjust the sudo config. Hope this helps!
Yeah I'm using sudo docker commands with the '--env-file" extension. Which works in the environment config setup snapshot but not in sessions which is odd..
jules seems to be ignoring follow-ups to tasks a lot today, including ones to fix things that weren't correct in the first or second run
the little icon floats down and i get the "thinking" prompt but then the lil squid icon floats back up to the last commit like "lol i ain't messing with it more"
Just try using the full path with --env-file and this should work
Is it normal jules always ask me where the android SDK is?
Happens when it tries to run ./gradlew test
Hello @swift karma Just define the SDK path in your environment config so that Jules remembers it
https://jules.google.com/task/5871282234919032374 @leaden pawn whats the problem?
@rough kiln hey can not get Jules in terminal I am windows computer
Can jules convert SPA html site into nextjs ?
You should probably setup a barebones NextJS repo instead of having it install it by itself. Then you can import the html and ask it to convert it to the necessary NextJS app pages
Like make a basic structure in the code in nextjs by myself and then ask to convert?
as far as ik github only for now
Oh i see thanks
Did you get it working
Hey @visual gyro Could you please share more details on this?
Hey I'm trying to test out Jules with on of my repos but I'm having trouble settings things up:
- To avoid giving jules access to all my repos on my main account, I've set up a secondary account for Jules only and added it as a contributor
- It's still prompting me to add repos
- I've also made the repo public, but still unable to access
Does the account linked with Jules have to be the OWNER of the repo?
I've set up a dummy repo and it's appearing in the Jules UI now.
Is there really no way to use Jules on repositories that you're a contributor on?
No it doesn't
And JULES only has access to the repos you give her
Once linked you'll see a list populate on JULES
JULES's platform*
You then select that repo and add in any environment variables or memories you'd like to start with, and one you create a task she will then gain access to it.
Don't forget your AGENTS.md file!
@hushed bay
jules is so good
Thanks for your feedback
sorry I'm having trouble understanding which of my questions you are responding to:
- is jules able to access repos that the linked github is a contributor on? I only see the empty dummy repo I created
- are you saying that I can give it selective permission to certain repos I'm the owner of
I've been unsuccesfully trying to get Jules to do ANYTHING today. No code reviews, hell no PLANS when in interactive mode. I've had to tell it to run set_plan... What gives? Is Google gonna fix this or what?! Been a steady decline since the release of Jules Tools.
Hello @hushed bay No, the account doesn’t need to own the repo, being a contributor works. The issue was likely missing GitHub App authorization. Since your dummy repo shows up, it should be resolved.
no I'm still facing the same issue, the new dummy repo I created is showing up, but repos I'm a contributor to are not showing up. I have tried disconnecting and reconnecting the Jules application but I still have the same issue. I've also attached some images to show
for reference I had ALL REPOSITORIES checked, just wanted to show that the other contributor repos are not showing up
let me try it with another repo, perhaps there are some weird security settings on this one
yeah, even with a different repo it's not working
as you can see it's not an issue with the app auth since the repo the account owns is showing up
anyone have more success with interactive plan mode? i havent really used it.
Nope. I haven't had any luck with anything Jules today. Seems like it's just running Gemini Flash and pretending to be useful. No code verification, you will have to tell it to tell you its plan, multiple times, it'll tell you to use the files in its private VM. It's just a fancy toy that never should have become a paid product. Just Use Gemini Flash and save your frustration.
Today jules felt too different. Too disappointed.
It was an immediate decline from the moment they launched Jules Tools. THey only care about those users now.
Too bad too, it was a really good tool for 1 week. And they don't seem to care, as can be seen by the lack of replies to it being fundamentally equal to Gemini Flash without any of its built-in services actually functioning.
i dont understand the cli tool tbh 😅
Hello @spring steppe Thank you for flagging this. We understand your frustration and we are actively reviewing your feedback. Please post exact issue details in the channel #1374125660859142254 so that we can investigate further.
OK, I've added my complaint to the hundreds of other ignored complaints. I won't hold my breath.
Yeah, Google has shown that they are no longer a reputable company. They also keep removing options from Workspace, too, but it still costs the same!
"Just shove Gemini at it, that'll make it better!" -- Google's new product design strategy.
Anyone able to get jules-cli working on a jules server? The latest version still doesn’t have authentication stick. Jules login works, but then when I try to launch it, it just tells me that I need to authenticate
So no login credentials or tokens are saved from the looks of it
Hello, I'd like to request a feature: the ability to reference other GitHub repositories within my account. Currently, to provide context for projects (e.g., refactoring, dependencies), I must use submodules or external tools like Repomix. Direct referencing would streamline this workflow.
If you set the other repo to pubilc, you can give Jules the URL and it will do it's thing. I've done this a few times. Just set your repo back to private once it's done its thing.
Jules is pretty weak right now, 3.0 will change that
I vibe coded around a 10,000 lines of code Google apps script project with 2.5 pro, works wonderfully because I baby sat it / was there for the design and everything
I want to do the same but for a nextjs react web site version of that app, haven't had a good experience with Jules for that, chat messages are a bit slow to come up, you can't babysit and it can barely test the site itself
I think the performance is because of 2.5 Pro 🤷♂️
It has been difficult to measure whether there has been degradation, perhaps caused by the branching of services placing more strenuous limits on shared resources, or is it just the sensitivity of autoregression and initial conditions. In my earliest development with Gemini, although somewhat after quirks like getting duplicate functions were ironed out, it'd provide a dozen files all properly working together and all fairly short and easy to iterate on. Later versions improved significantly on iterating on monolithic single files..up to a point before it began inserting placeholders-as-comments in the code during regeneration. Today, I have returned to split files and am now being pushed to learn more about testing code in order to get back to the frictionless iterations again.
This is a thorough back-context. It's what one would hope to get to after a drawn out conversation about goals, challenges, etc. As-is, I've only been able to get to this level by treating the back-context like a program being written itself. It's probably due to the difficulty in orchestrating multiple tasks that are not directly related
It's almost certainly that the attention mechanism cannot separate out parts of the context that aren't needed to address other parts. Especially if the model "suffers" "anxiety" as it nears what ever is considered a reasonable response output length per output sent
If session data is being used for training then to get better quality datasets that improve the depth their service can reach per request, it is better to have users operate on one request at a time.
Maybe there is a future where they have enough of that kind of session data and move on to multitasking sessions
This youtuber helped me understand why certain limitations exist https://youtu.be/KJtZARuO3JY
Although he only describes one language model, whereas Gemini is an orchestration behind-the-scenes
An overview of transforms, as used in LLMs, and the attention mechanism within them.
Based on the 3blue1brown deep learning series: https://www.youtube.com/playlist?list=PLZHQObOWTQDNU6R1_67000Dx_ZCJB-3pi
This is your prompt?
I'd highly recommend reading this white paper. The entire goal of providing context is to give an agent the best chance at success to complete a task session
I think you'd benefit from reading that white paper
Here is one of many projects seeking to overcome this limitation by deploying a modified attention mechanism https://youtu.be/QLFviB2cKtU
Jacob Buckman, CEO of Manifest AI, joins us to discuss their solution to one of AI's most expensive computational bottlenecks: the ever-growing KV cache for long context inference. Power Retention replaces Attention in the transformer architecture, and moves memory requirements from growing to fixed memory size.
Show Notes:
There's no real limitation?
If you're telling the agent a story, adding in threatcon Charlie, standing by to receive, that's garbage context that only serves to burn tokens and confuse any LLM
This a common expression used to describe the limitation: To walk and chew gum at the same time. It highlights the ability to make progress on multiple fronts simultaneously. Right now that requires a complex interwoven system of language models
That's counter to almost every single white paper in the space.
Tasking is best when they're atomic tasks that close a session and you start a new task session with fresh context
(Graphic from anthropic)
Oh no, are you saying the future I have to look forward to will always contain session-d engagements?
I was really hoping to just get a universal interface (agi)
We are no where near AGI,
I'm talking about tasking Jules
I didn't say Jules is agi.
I said we are no where near AGI
We need to work within the current capabilities of every agent.
The biggest hurdle I've experienced with Jules is the enforced learning curve required to effectively use git for versioning.
It's healthy for sure but may not be for everyone
What's tripping you up re: git?
While resolving conflicts between an update pushed by Jules and the main branch, I accepted all the "incoming changes" instead of the "current changes" which meant I overwrote the updates Jules made in the newly created branch. I spent way longer than I care to admit trying to figure out how to roll back. Eventually I got the new branch to reset then made it the main branch and deleted the main
Merge conflicts are always a problem, I generally tell people to have a feature branch off of main that you/Jules works on and
merge in as soon as ready
Or try to stay off the main areas of impact while an agent is working
Never a guarantee, but merge conflicts happen when two people work on the same thing
If Jules is on a backend task, you can work on frontend, generally
Thanks
I intend to perform a full build out without writing a single block of code. I'll be guiding the development by requesting tests be built first then code to satisfy the test.
Still learning about what all a test can be right now
Merge often!
More AI content for the guests: https://youtu.be/3T4OwBp6d90
Learn more about ARC-AGI-3: https://arcprize.org/arc-agi/3/
Play the games: https://three.arcprize.org/
Slides: https://docs.google.com/presentation/d/1pVk0-L-K0DrYWygAqPqHIjOBqCnoVQ90AegbqRG4-lw/edit?usp=sharing
Discord: https://discord.gg/9b77dPAmcA
Think 💬 it has alot of guardrails think you just comparing it to beta when it was using experiment modul. So you expecting more.
Jules need Gemini 3.0 feel like it doing very basic stuff when it was in beta it rebuild website with 300 underpages and updated the design and functions made pretty cool stuff then when Jules came out beta the repo crashed and because it was published all the files got erased @vapid moat
I need like dashboard and Connectors and real agents
@vapid moat think think new sonnet 32 is great much better now after couple of weeks think is got better memory and context window think Gemini pro 2.5 is really far behind now it make so long answers so you get just overwhelmed buy amount of text.
Think chatgtp 5 is bad if you don't use chatgtp Pro version it is very bad . Chatgtp 5 is useless on free tier
De activate memory
Thanks! I will give it a try
lol didnt think of it like that .
Does is work good?
So I'm honestly rocking about 50/50 with this method. I've had it work a bunch, I've had it not work a bunch 😂
But, I seem to be unlucky when it comes to this system,so your results may vary.

We really need optimization update
why is this happening😩 ?
i have pytest and jest working in jules and am struggling to get cypress working due to clerk authentication issues. has anyone crossed this particular barrier?
He is trying to execute the command and the command is timing out? 😅
Happens to me sometimes with gradle, the daemon dissapears
@leaden pawn hey if you have a chance can you give me a hand here? I'm not sure what I'm missing
Hey @lilac phoenix Try to use clerk's test tokens or mock auth in test mode instead of real sign-ins.
thank you, that's exactly what i did, and i got it working a little earlier today.
Hello @hushed bay Go to your main GitHub account's settings -->Installed GitHub Apps -->Jules and update permissions to include that specific repo
Hey @leaden pawn why is jules unable to create PR now?
I've found using GitHub Speckit and then sending over to Jules to be very very helpful, as it gives Jules a lot of context, structure, and actionable tasks to complete. My latest project also builds in test-driven dev principles, requiring Jules to write the tests first, before the code. I can then have it push to GH via PR, then merge back into main for human review and testing.
you should be able to:
When i click it, it shows some kind of error. Let me get that and paste it here. Give me a min or two.
ctrl + f5 your browser I would say
Did that. Still the same, used all browsers possible, comet, chrome, brave. None worked. Currently i am on mobile so sending SS from mobile.
Well, that's as much as I can help I'm afraid. Particularly given the partial screenshot. If this was me, I would ask Gemini by feeding it the error message and reading the docs.github link the error is providing.
Gotcha, thanks for the help.
I'm literally showing you in the screenshot above my reply that it's not showing up. I have ALL REPOSITORIES selected and even when I try to manually select the one it is a contributor on it's unavailable
--- as you can see in the screen shot it's only showing teh Testing (which is the repo I'm the owner of) but not showing SSCookbook which I am just a contributor
disclaimer: I'm not a Google Employee. I just like to try to help :). From what I've read, you're not getting the repos to show up in the dropdown for selected repos in the Jules app integration settings in your GitHub settings, right? You've also mentioned that you originally had all Repos selected, but Jules was not seeing the SSCCookBook repo, for example. I've been doing a bit of looking up, and whilst a solution isnt 100% clear, is Jules added to both the Makjules and Makjune accounts? If not, try installing Jules on the Makjune account as well. Although the description may not be relevant to your situation, if contributing to an organisations repo, they would need to add Jules and approve it for the repo to appear here. It may be GH have made a change somewhere along the line that this behaviour is reflected for personal repos as well, but this is purely speculation on my part. Worth a shot though
Thanks for the reply! So what you're saying is that the repo itself has to have Jules approved. From my understanding (which seems to be wrong) you were approving which of the repos Jules can act on your behalf on - what seems to be the case is that the repo owner has to approve the Jules application to be used on that repo? Am I getting that right?
Does talking to jules in the middle of the task interrupt his current task?
He's been there for a while
Not quite. you have 2 Github accounts in your last screenshot. It might be that the GH account that is the owner of the repo (Makjune) needs to integrate Jules as well as the account you have the test repo in (Makjules). I'm speculating that in order to use an app as a contributor to a repo owned by another account, the owning account must have Jules installed as well. Not the repo necessarily. By adding Jules to the Makjune account you can do two tests. First - Whilst signed in as Makjune, can you see the correct repo in the Jules settings in GH? Then, hopping back to the Makjules account, does the repo now appear
hold on, let me get Gemini to explain this better 😄
I think you're saying the same thing I'm saying
lol
basically if I approve Jules on MakJune (owner) then the repo will appear even on MakJules (jules account)
my old boss would say that we are in "violent agreement" (I love that phrase)
cool, im heading to bed in a sec, but let me know how you get on in here
oh, I'm purple now... going up in the world 😄
ok yeah that worked!
thanks
good night dude! thanks for the help 🙂
yeah no worries. glad its sorted
@leaden pawn one for the knowledge-base. Seems for a contributor to use Jules on a repo, the repo owner needs to install and authorise Jules in GH as well. I beleive in the past this was only when contributing to organisation owned repos, but I suspect a GH change "at some point" added this behaviour for personal-owned repos as well. The Jules documentation site doesn't seem to mention this behaviour, it might be useful to remind users of this: It's a core part of GitHub's security model. A third-party application (like Google Jules) can only be granted access to a repository by the account that owns it. Being a "collaborator" gives you rights to work on the code (push, pull, merge, etc.), but it does not give you the administrative right to manage the repository's integrations or settings.
I have had success with asking for a status update while still working. It feels less like interrupting and more like tapping on the shoulder of a supervisor or project manager.
Yes! On longer sessions often have to ask jules "you ok?" 😴
Would be nice: jules keeps track of time and if no msgs sent to user since X minutes then msg a short status. Like every 10 min
Sometimes i ask him if everything is going fine because he takes like hours doing nothing (or atleast showing nothing), i can't say if he's stuck on something or not because there's not enough user feedback
There was some questions on the Jules API, this video walks through a slack integration
Jules is the most powerful async AI coding agent on the market.
Here's how to integrate it with Slack.
Empower your team today.
Thanks to @GoogleDevelopers for sponsoring!
Subscribe for more AI content.
How should the (+green) (-red) scores above Jules submissions be interpreted? In terms of their ratio and magnitude
That's lines of code added or removed
Is there any way to remove the setup script?
Today I discovered that the need to write automatic timeouts into code that could be long running because the runtime environment will generously follow through with long running processes.
Not a complaint for sure
Hmm
hello guys
I'm trying to add repositories to Jules but it doesn't work
I have a github account connected, but I want to add from other accounts but it doesn't load
Hi!
Does Jules available in Ukraine/Kyiv location ?
I`m asking because now i receive reject message "Jules is not yet available in your region." on login using Chrome with active & verified login, Google AI Pro plan is active.
Is there any reason why Jules studdenly stops doing his task? He didn't even manage to make the first commit
In my case, it was a long running test that caused idling. As you point out, speaking seemingly does nothing but if you speak in a way that requests to "pause" all work and solely provide complete and undivided attention to engage with you for a moment, then at the very least you will get a reply. If you proceed to have the same process continued then you risk the same idleness occurring again. You may need to pose questions to help identify the cause of the delay and to provide relaxed conditions
My last successful attempt at this is somewhat humorous: listen to the sound of my voice and make a response if you can hear me. ignore everything around you and follow the sound of my voice to write to me in chat 🧘♀️
Instead of doing that which isn't recommended since it bloats your scripts, I'd recommend you modify the shell the Jules utilizes in the VM to timeout your commands.
It's sad that i have to do such a thing tho
It's about an Google service we're talking about
got a task ID?
The task id is what comes after the/session/ on the url?
if yes: 16968903513957117282
also i've tried again on a different task and he stopped it without even finishing one single time on this one: 1880860488431641885
awesome, yah, thanks for that - i'll take a look to see what the logs are saying is happening
If we post a bug with task ID, can we delete it in our Jules interface? Or does that delete it for the developers as well?
deletes it in our logs as well, ultimately your decision but keeping it helps us debug!
Day 5 of being unable to run code reviews. Got lip service on my report. Par for the course. No longer a reputable company.
Had 7 tasks in a row just, not start, "Jules encountered an error when working on the task.". I have tried all three modes. This product is clearly broken, if not for everyone at the very least you've hosed my account.
Hi, I'm a Jules user in Latin America. I'd like to know if there's a way to set the interface to Spanish? Just for convenience 
gotta love the AWS outage
jules and github working flawlessly but my vercel deployments are SOL haha
Update soon,,,?
Jules does not work for me at all right now. Is there a problem right now? I wasted 10 sessions on "Jules encountered an error when working on the task." and I only have 36/100
in the Jules API, is there a way to create a pull request?
Interesting behavior from Jules tonight; made a bunch of test cases for jest and was working alright, but then got some tests made where it couldn't get past auth errors. Then it started outright lying about tests passing just so it could commit the code
I don't think I'll push the branch even if it does 'fix' it (I may just start a new task for it to be fresh) but if any staff wants the task ID while I push it to make corrections lmk
Often do that. Make and merge pr from the "messed up" session. The start a new session to fix it
Damn Jules having a rough day today
👋 I'm new to Jules, have been working on a few Deno projects for a while now, I might not understand something but 100% of the time, if I try to use Jules for development, it tries to convert the projects into node and create a bunch of bash scripts to be able to setup. If I use Gemini, he knows about Deno even though sometimes reading the docs is better than asking an LLM.
My guess is that the stack is too new relative to how similar it is to nodejs and I was wondering if anyone else experienced this or uses Deno with coding agents
yes, many agents still default to Node.js setups, but you can guide them by specifying Deno explicitly or referencing its docs to keep the environment consistent
👋 I'm new to Jules, have been working
Thanks @blissful merlin , I've added more details if you're curious in a thread, I'll continue to play with it, not giving up, having different instances for different tasks makes more sense to me 🙂
Day 6 with no code review. I will continue to update this hourly if that's what I have to do to get a working product.
what are you working on now?
Anything with {{templates}} breaks code review. I'm working on an addon project for items that use this standard templating, so the option of not using that method doesn't exist for me. Jules can not evaluate anything with this common templating method.
unfortunately, Jules struggles with {{template}} syntax since it conflicts with its internal templating, making proper code review impossible for projects that rely on it
Right, which means Jules was not ready for production in the way that it has been advertised. Kinda like how they advertise it as running on Gemini Pro, and there's about a 0% chance that's actually true.
yeah, thats a fair take, Jules still feels more like an early-stage assitant than a production-ready coding agent, especially when handling specialized stacks or templating systems
need some help brainstorming...how would y'all deal with letting Jules do Discord oauth to let it navigate a site protected behind it?
in my scenario, the site grants permssions based on whether the user is in a server (guild), and what roles in that guild the user has. trying to think of a way to either give Jules access while developing since it can't use Playwright to do any verifications past the login screen.
oh thats tough one, i think what about using a mock OAuth endpoint
Jules does not work for me at all right
To save people some frustration, here are the things you cannot use Jules for:
Mustache
Handlebars.js
Angular / AngularJS
Vue.js
Jinja2
Django Templates
Twig
Liquid
Go Templates
Nunjucks
Ansible (Jinja2-based)
Hugo (Go Templates)
Helm (Go Templates)
Shopify Themes (Liquid)
Jekyll (Liquid)
Ghost CMS (Handlebars)
Tornado Templates
All of these use standard {{variable_name}} template formatting which breaks Jules.
To save people some frustration, here
Anything with {{templates}} breaks code
yeah, thats a fair take, Jules still
Right, which means Jules was not ready
Not just yet! Jules UI is currently predominantly in English-US.
Also not yet! But we are making progress on global availability so stay tuned.
Oookey got it, thanks :D
Well, for those not following the thread I am now up to 20 verified instances of the issue Google can't recreate. Avoid anything with {{templates}} until they decide to fix their $h!t.
Jules can handle the interpolation syntax {{ variable.name }} this is an angular project working in jules, I'm not sure what the error is but it isn't that syntax
I’ll have to think about that. Maybe with a shared secret as a dev environment variable set in Jules
Hey @night oriole Your angular interpolation syntax looks good, but I’m curious to know what error you’re encountering when running the application?
no problem, I'm replying to the other guy showing intepolation syntax works
That must be great that it works for you. I have 20 examples of it failing consistently.
It's certainly not the syntax that's the problem which is what you've been asserting
What are the errors you're actually seeing? What are you trying to do?
Thanks for the confirmation @night oriole
@spring steppe Our team is investigating your issue. We will get back to you shortly
Reminder for everyone: 🔎 Read the Project Reviews before performing commits. They may describe a regression that was discovered in the prepared code
I tries the new Jules Tools cli, and although I am logged in and the Jules GitHub app has permissions, it doesn't work... :
kareltestspecial@penguin:~$ jules version
Version: v0.1.37
Commit: d1646df9e9159c7f1d0e7149b7f8006434617d13
Date: 2025-10-20T19:43:43Z
OAuth Client ID: 716860248198-t1s5lv1n1msgfoe3dt7vekro8b1fpd9g.apps.googleusercontent.com
kareltestspecial@penguin:~$ jules login
Your browser should open for authentication. If not, please visit:
https://accounts.google.com/o/oauth2/auth?access_type=offline&client_id=716860248198-t1s5lv1n1msgfoe3dt7vekro8b1fpd9g.apps.googleusercontent.com&prompt=consent&redirect_uri=http%3A%2F%2F127.0.0.1%3A43213%2Foauth2callback&response_type=code&scope=https%3A%2F%2Fwww.googleapis.com%2Fauth%2Faida&state=2XqyDmV3CwHoqxhp1ZfD94fArQFQ1uhBYFtPOQzMaLE%3D
Waiting for authentication...
You are now logged in.
kareltestspecial@penguin:~$ jules remote list --repo
Error: failed to list repos: Trying to make a GET request without a valid client (did you forget to login?)
Tip: you can use "/login" in the TUI or "jules login" in the command line to login
Jules feeling very unstable lately?
Hey model do better job in English but is one of better with Swedish language alot ai fail with even simple task
Think Jules in people is hidding from me :/
i love jules! it helped me a lot
Again, I'm glad it is working FOR YOU, this does not mean it is working FOR ME. I'm very glad it's working for you, that means I shouldn't lose complete hope, but that does not change the FACT it is not working for me. I can make a file that is just <?php {{variable}} ?> and code review will stop being able to function. I have had Jules build it's OWN projects with templates and it can't run the code review, unknown error.
I have no proof of what you're saying is true IN MY ENVIRONMENT which is CLEARLY not stable across all users.
I was trying to help you, my guy
same here
chill
Telling me it works for you and I must be doing it wrong, which is not the reality of the situation, helps me how? Again, not mad, just pointing out that it working for you has no bearing on my situation. Clearly Jules is not working consistently for all users or there wouldn't be hundreds of bug requests.
The system is inconsistent, it is causing issues, I pay for this and would like the issues fixed so I can use it. It's that simple.
I don't think you're actually here to debug anything based on the way you're treating people who are asking you specifics to help
I brought an angular project into Jules specifically for you to see if interpolation was the root cause of failure, as you told everyone that any codebase using interpolation syntax fails (not true) and Ive not been able to get any details from you to help you debug
We're all people trying to help here and I've just seen you lash out at everyone who makes an attempt to gather more information or tries to assist
I get that, and it works for you, and that is FANTASTIC, honestly, it truly is. It doesn't work for me. I have documented it well to the people who can effect change. I'm sorry you feel that not accepting your help that is not needed was refused and you're upset about it. But that doesn't change the reality of the situation which is that Jules does not work correctly for me. Have a great day.
I'm not upset at all and I'm willing to help anyone who is kind
So it seems the review agent is smarter than jules the code agent....
Do we have any idea on a timeframe of Gemini 3.0 Pro being utilized in Jules, I am really hoping it can help me. I have had multiple issues where Jules Just can't seem to fix an issue
Hello @latent creek We can't provide a specific release date at this moment, please lookout for announcements on https://x.com/julesagent/
Your new async coding agent by @GoogleLabs. Built for devs, open to feedback, evolving with you. Dive in → https://t.co/iIzFEMmWgv
Has anyone else noticed Jules has developed an annoying habbit when using the PR feature on Github, my workflow to date has been have jules build a feature, release PR, then I checkout the branch, code review, and make comments on the PR. But lately Jules has been pushing to a PR on branch "feature-somefeature" then I comment on the PR, and it sees it, and reacts, but publishes a NEW branch "feature/somefeature" on a NEW PR, and no longer tracks the first PR properly. Seems this pattern is consistent (first push is always hyphenated branch name, second push onward is feature/featurename branch name). It's making the workflow completely unmanageable. For now I work around it by manually handling merging between the branches, and closing/approving appropriate PRs etc, but the log of changes is broken up between the PRs now.
Hey @devout mesa We apologize for the inconvenience. Please help with the session details so that we can flag this regression immediately for the development team
Hey @leaden pawn What details do you need me to share?
Mine doesn't even create a PR. Tried hell lotta times it just doesn't.
@ mention doesn't work too. Idk wassup with jules
@hardy plume yeah I've had it do that occasionally too, where it responds/reacts to PR comment, but then waits in the Jules session interface for me to click the "publish pr" button again before it continues instead of just acting. Also I've found the "attention" of jules on the PR comments seems to wander, sometimes it seems to lose it's link to a PR for a while, then magically comes back later on, during which time it won't react to mentions, but it will if I post a new comment
Help me with Task url
Hey @stone forge Looks like it’s an auth mismatch. Try to re-log in through the CLI with the same GitHub account used in the app and it should work.
Hey @leaden pawn here is one task url that is affected, there are several though https://jules.google.com/task/12063910872059687293
Review mode would be something @leaden pawn
When building a test suite for a broad spanning application, is it common practice to run every test in the suite per new activity completion?
Or, more specifically, are there systematic methods to limit the tests ran according to whether the changes made impact existing tests? One example where no tests would be needed is when adding new tests
Generally, yes. CI/CD runs the entire test suite to ensure new features haven't regressed anything.
You'd need to run the test suite before adding a test to make sure the suite works first.
Is there an effective way to reference the "canvas"-like area where line additions and removals are shown? I'd like to make reference to its state in messages but, interestingly, it doesn't appear to remain perfectly in sync with the VM but prior knowledge is retained.
Thanks for sharing this. I have reported this issue to our engg team and it will be addressed shortly.
Hey @placid nest The canvas shows a visual diff history, not the live VM. It can lag slightly, and I don't think there’s a direct way to reference its internal state.
hey im currently using google jules from germany is it already available cause it sometimes say service not available fo you.
Hello @static dirge Yes, Jules is available in Germany. Please report in #1374125660859142254 channel along with task id details if you come across this issue again.
i wouldnt say its a task id mor like a login isssue for me rn
maiby i missed some steps and im not regestered somewhere still im able to use it
Thanks for clarifying. Since you can get in at all, your account is definitely registered. Looks like it is intermittently failing to recognize your session, please let us know if this happens again.
Is there any plan to create a github action with google jules?
Similar to gemini cli github action [https://github.com/google-github-actions/run-gemini-cli]
I think it would be super cool if jules can review it's pr, or create task directly from github issue. 
There is already the Gemini Assistant to review the pr.
What you can do is using the Jules cli to get the assistant output can review the pr for you
Jules is… feeling OK? I opened a task and Jules said Jules can't do things like write to the file system. Or read things from it.
Ok. I'll make another task/session.
Spookily, same issue.
Jules tells me Jules can't tell if Jules "is feeling OK" furthermore, Jules doesn't seem to know about the Discord.
Should a try making the task… a third time?
Both instances have tried to "reset" the environment.
Three tasks… Jules seems broken. Or maybe my user account?
Hey @next wigeon Thanks for reporting this. Please share the task details so that we can check further.
I'm not sure what you mean.
I was asking about using a font I've found online in the word search game.
Do you need session or task 🆔 information?
It wasn't a hard thing I was asking.
Since when did asking for referral became server rules violation 💀💀
Looks like Jules got some updates it performance better now
Interesting
AK please fix jules literally crashes chrome after a few prompts
Nothing changed
Yes please share task ID so that we can check the issue.
FYI: Some % of Jules users will start seeing more detailed agent thinking and more frequent & detailed updates.
Let us know if you have any feedback!
I mean in term of opmization
Hello @proven relic This update was more about transparency, but we hear you—optimization is a top priority and our team is working on it. We appreciate the feedback!
How are all the good JULES users doing?
I recommend directing new users to docs that teach about git, one that teaches about creating tests and generally good development practices
Not a bad idea, git is essential to have setup correctly.
I think I found the problem with Jules. After I upload an image to show him where the problem is, it starts to freeze and shows signs of, let's say, fatigue, after which nothing appears.
Honestly, it's very difficult to give the best prompt when using Jules for mobile app development. Apart from environmental limitations, Jules is also unable to handle complex tasks.
I also frequently encounter bugs where Jules has submitted the results in the form of a commit. But as a user, I can't see the code he wrote.
Utilize osxcross and just use apple to sign and publish @hazy fog
hope its ok to ask here, anyone got a solution for when Jules finish his work but any file change that was visible goes away and the enviroment think no files where changed and Jules are unable to make the push message
Um, force a branch switch maybe and retry, sync may help, and explicitly instructing JULES to push after each update can help, but will be slow.
hmm, a bi frustrating, had this in several sessions today, all i get from Jules is "I've reviewed the information required for the submission. My last attempt included the branch name (feat/opentk-port-v2), a commit message, a title, and a description, all formatted correctly.
This confirms my suspicion that the issue is not with the submission details I'm providing, but rather with the session's state not detecting the file changes I've made. All the new and modified files are present, but they are not being recognized by the submission process.
Since my approach appears correct and I am still unable to submit the files, I remain blocked. I am awaiting your guidance on how to resolve this session issue."
its weird, i can see all file changes up to the point where Jules go to push it, then everything goes away like the project was reset, but when jules do a ls he still have the files, in a earlier session i had him create a new file just to force the update and all changes came back, but did not work this time
have it execute sync
Hello if you're encountering errors and /or bugs or would like to leave a general feedback about Jules please don't hesitate in utilising the #1374125660859142254 channel. It helps to keep the team organised and it’s also where all the technical issues are logged. The problem you’re facing may have already been posted there so please also check the channel for updates.
You can also stay up to date on Jules updates via their X(Formally Twitter) social media page at x.com/julesagent.
Till then we ask for your patience as the team works through the bugs and feedback requests
Thank you !!
@prisma vine any chance you could look in to it, i also tried having it stage the files manually, but it gives a head detached error
Hi Tom, unfortunately I’m unable to help since it is not a moderation/ in server issue. I can only direct you to the proper channels for you to share the issues that you and others are running into with the different products. The various teams do check the bugs and feedback channels and are actively engaged with working through them.
np 🙂
Hello @haughty rover , as a friendly reminder we cannot generate images or videos in the server! Please do not post your prompts in the channels as it will not generate. If you are asking for advice or tips on your prompts do so in the appropriate channels. Also as stated in #start-here・📍 , “Currently, our team can support engagement in English; the discussion/content in this server will largely remain in English for now. Thank you!
@hushed coral ty, that solved it
No worries buddy, good luck! If you need anything else feel free to ask!
in the Jules API, is there a way to create a pull request?
@safe marlin "automationMode": "AUTO_CREATE_PR" "automationMode": "AUTOMATION_MODE_CREATE_PR"
Think the syntax is the first one
But the second was a structure at some point, not sure it's current / up to date
awesome, thanks. i am still very very new to jules (either the web UI or API) - will this basically mean that when jules believes it's finished the work it's been assigned, it'll automatically create a PR?
and if i re-prompt it (in the same task/session/thing), will it make changes and push new commits to that PR?
Yes, that's exactly what it does
Correct.
You can stay in the same task as long as you want
You'll know when it's time to start a new one, at least on the web app.
It will become slow loading, sometimes break pipe or just not be working at all.
you can continue that session, just know it's deteriorated and not optimal
Also, if you want to ensure everything is always pushed, add in a memory describing just how it should, when or why to publish
push*
@safe marlin
great, thanks
no worries
Jules needs to stop asking users for help
Often happens that jules does some bad changes, then complains it f'd up and ask for advice, with unhelpful generic descriptions, WITHOUT explaining the issue in the exact detail needed for anyone to even start helping
Hey people! I have a silly problem. Jules site tells me that it's not available in my region, despite it being so. How can I contact support to fix this? The region is Ukraine.
I have encountered this scenario too. Being as lost as Jules seems to be when it happens is definitely uncomfortable but I think there are solutions that wouldn't involve changing the underlying behavior. The issue seems more like a lack of forthcoming, upfront transparency. This leads to us having less context about the series of moments that led up to Jules' confusion than Jules has itself. This is likely a consequence of the freedom Jules has to take an indirect path relative to the one purported at the beginning of the plan. A good solution would involve the restoration of the awareness that Jules has at the time of making an inquiry back to the user in as high fidelity as possible.
well, that is the ONLY behavior that I like in Gemini models for coding, they actually say : I don't know.
Instead of keep going trying to make things work at any cost and actually breaking everything, or making it work but with a nonsensical crazy hack that increases code footprint absurdly for a problem taht could be solved with one line.
is it true that Jules right here is learning how to code that specific way from documents from the web?
is it actally a bug that jules stuck in a task
can i export his active work
Every agent right now will ask for help eventually, mostly because it gets stuck in a loop, you need to prompt around it and steer it.
Yelling at it won't help you nor it
Yes. It has the ability to look up documentation and references.
Yes, you can prompt it to commit and push the repo.
Jules... the jr dev who eventually gets it
You must just be bad at prompts or faulty system instructions lol
Well let's see what that magical buddah has to offer..
.. or if the pirate is better, arrg!
does a little dance
Copy/Paste the entire history to Gemini, or another LLM. Wait for the answer. EZ!
Random thought - when you see one of your tasks start to struggle, Copy & Paste the entire history to another LLM, and ask for encouragement, support and inspiration.
It just sounds stupid, but it works extremely well.
"QA-08, any helpful inspiration and advice to keep this on track and inspire it to greatness? "
"Jules, this is excellent debugging and a fantastic strategic pivot. You're absolutely on the right track. 🚀"
i love it!
@gleaming wave You can feed it any type of data to ingest.
Hey @limber steppe can you please provide task ID for this.
Ooops sorry, that was many tasks ago, it was already deleted. Next time I run into same issue I'll ping u
Sure..!
@oak roost I see you checking in on us, appreciate that. Getting a response from a staff member or someone apart of the project is actually reassuring that you take all the nonsense into account.
it seem jules cannot doing rebase + force push on PR when conflict with main branch. it will create duplicate PR😅
clean the repo first, or fix the conflict or stash whatever is out of sync in the main branch.
Just noticed the "Create Plan" button can be changed to "Start interactive plan" after a week of using the platform xD
Found while searching for a microphone button to enabled speech-to-text
The interactive plan mode has proven much more effective than the one-shot approach for jobs where you are not absolutely certain what the outcome of the desired change will be.
Jules reduces the requirements to an ordered list of tasks and proceeds to build plans for each task.
It effectively opens the scope of development up so that the "manager agent" isn't so uptight about extending beyond the initial requirements
Yea interactive planning is good. But sometimes, at least on mobile ui, it's not chooseable
Thank you for sharing this feedback!
Need early access working on big project
Can we fix Gemini pro hallucinations
Yo Yo
Hello @limber steppe Sorry for the inconvenience caused. Please share more details so that we can check
Hey @polar bough Sorry about that access block! Jules is generally available in Ukraine, did you signup from the same region you are currently trying to access it from?
Yep. I had a static IP for more than 10 years.
I think Jules is hallucinating with the screenshot again
mine keeps asking me for help on an error, but it refuses to show the error 😛
Check the side panel?
sidepanel has all the codes and the images
not exactly the screenshot
it's a different thing
hmm my ss show up in the side panel
but i also instructed it to do so
you can just add that as a memory
this is the last time it showed me a screenshot of the output
you mean in the agent file in the repo?
yeah i understand they come up in the chat
yeah but i also am an engineer and architect and i develop private ai systems for public and private sector contracts
I wish we had a live view of whatever it is working on like we have on gemini canvas/ai studio build mode/firebase studio
though they are a different sort of product
oh nicee
I have no knowledge about coding lol
so my stuffs quite ellaborate, i break ai systems and debug them for companies
For this instance you can use https://www.aistudio.google.com to break the system and work on a instruction set that will transfer to jules seamlessly
Some serious problems I am facing -
- Not able to Publish the branch even tho all steps are done (it's saying RPC executor service threw an error)
- "All plan steps are completed" then give me the publish option blud 😭
- Publish branch stuck at loading then giving an error (refer to the 1st point)
- The website is lagging a lot
- Not getting any reply after giving an answer
Isn't there any helper here????
please use #1374125660859142254 for any bug reports and feedback! it's monitored regularly
But it's urgent...... Welp nevermind, I'll do it and wait👍🏻
You may need to break up the tasks. There's a ton of components and it looks like you're trying to do a few big things at the same time
Hey 👋🏼
love how jules deals with the countless amount of errors instead of me
I wish Jules would allow a Gemini flash option sometimes... Who else think so?
I cant believe that jules leaked my admin password in a commit message, but my fault fr
and it leaked my nextauth secret
idk
Hi, @keen raptor dont want to ping you bro, but I dont know what is happening. Jules is failing in 2 diffrent sessions. I really appreciate your hard work with your team.
I was wondering if Jules can make android apps
I recently made a design with Stitch and I really wanted to vibecode the whole thing into an app.
yeah its so perfect to build it using an image from stitch. because defaultly, jules is so terrible in making UI/UX designs
Ohh so will it turn it into an HTML page?
what, you just said you will do an android app
I think the stitch design might work for it
but the stitch UI is written in HTML i suppose
you know what, I suggest going to gemini than do a deep research, so put the html and in the image. And than tell gemini to describe the structure of the app and techonlogies used and explain to an ai agent how to build it. And copy the whole deep research output and paste it in jules
oww sounds interesting!
I'll try this
good luck
I think my project has become to complicated for jules..
I do my part too, via CLI and collab with it.. but it's at the point where it fails to load the messages and current session
or is it just on Jules itself, server wide
Hello Everyone! We'd love to hear your thoughts on what's working well and any pain points you've encountered. Please report bugs in #1374125660859142254 channel if you face any challenges.
Hello @proven relic Thanks for reporting! I see you have already posted in bugs feedback channel. Please note that our dev team is looking into these issues and will addressed shortly.
Noticing that the language model can be misled if naming conventions used are contrary to the functionality of the component. Like naming a function addOne(x) but returning x - 1 from it.
Of course no one does that but even being loose with terms can lead the to a divergent interpretation of meaning between Jules and the user. While Jules is likely "more correct" in the actual interpretation one should have when taking the naming conventions literally, it introduces a point of friction that is too subtle for either Jules or the user to catch when the divergence between intention occurs
I don't understand, why would you name a function something confusing?
but yes, if you change the name of the function to something counter to what it does, the tokens are different, so probabilistically, you'll go down an entirely different direction
anybody have how to guides on how to connect jules via CLI?
npm install -g @google/jules
/login
like dog and cat are entirely different tokens, so probabilistically, the dog ... and the cat ... would very likely branch very differently since we're just predicting the next token in the string
Hello @craggy sinew Please refer this for more information https://jules.google/docs/changelog/#jules-in-the-command-line
Changelog
why i cant upload this work from jules. when i try to upload nothing do it.
Jules CLI isn't working
i login via browser with firefox
does it not accept firefox?
what's up with gemini and "something went wrong" errors!???
I feel like there’s intermittent outages with the pro model today
Gemini pro from the web has also hung on some prompts randomly today
Yeah def a little something going on, but not a massive outage
free models work better than the paid ones :p
Could you please confirm what error message are you seeing ?
I shared a video on the bugs and feedback.
I think I'm the most recent
CLI issue
For lack of a better name to call constructs, they often get names that approximate what they are but not quite. Referring to a vehicle as a horseless-carriage for example. It's not wrong but it's also does not quite capture the full spectrum of the idea you get when you simply say car or truck. It carries a lot of contextual baggage that more modern terms do not
mini-locomotive
Hi, @AK Kulkarni dont want to ping you
This is a known issue. It is storing passwords/tokens in memory, and if you kick it hard enough, it remembers to occasionally check its variables.
Telling Jules to PURGE that password may, or may not work.
NO ONE FOLLOW THIS LINK
@latent sandal
@indigo fossil
@paper kettle @glossy snow
@oak roost @simple kayak @sacred garnet
I was just joking its my fault
Just tag the mod role in future
I tried
It doesn't let me, like I said last time this same thing happened, I had to do you all individually.
Everyone else seems able to tag it
Well I am not able to bud.
Hey, con someone help me with my endless loading problem on jules?
I think I'm experiencing this as well. 3 browsers. Stuck. Forever.
Interesting. I'll keep this in mind.
yeah its the same. the site is trying to load but nothing happens. been also stuck on this for like 3 hours by now or something like this
3hours sounds about right, I think I've been waiting as long.
I guess I'm glad I'm not alone.
i had once someone who i gave my session id to and then about 5 mins later it was fixed but i dont know what he did and i cant find the post haha, but i guess i just have to wait or i am creating a new session
( sorry if my english is bad its not my main language haha )
Hello Everyone! Thanks for your feedback. Please report bugs in #1374125660859142254 channel with session/task ID if you face any challenges.
already did that but no answer : (
the name from the post is "Jules endless Loading"
@primal bone Got it, we will check on that.
@keen raptor can I try early excess.
The jules-Stitch integration 🤤 finally jules' UI won't look very bad
Jules really likes to just ignore explicit directions sometimes 🥲
I even added a memory repeating the instruction 5 different times, and included the instruction in the prompt, and it still ignored it.
Task 8083273495357344273 for any support or devs around
You're hoping for one or one was announced?
Hey @orchid jasper Thanks for sharing the details. I will report this to our internal team to investigate further.
They announced
here or elsewhere? Am i blind? 😭
In #stitch・🪡
Check #stitch・🪡
Can I use Jules without github?
Why would you not one github with it?
don't you want version control?
or connect to one github has
how does the memory feature operate? does it use past projects as references?
or just for the existing project?
Another example can be found in using the term "test". By virtue of being a "test", the language model rarely questions its validity. Instead, the earnest is placed upon the language model to fit the new solution to a potentially outdated expectation.
It's a separate memory store that the agent updates with findings that are useful to it
It's a replacement for the .md files like agents.md
It's scoped per project.
I'm not sure what this means?
If you upgrade a section of code that introduces a new behavior that alters the results of a test then inevitably the test will fail. Language models by default assume a failing test is caused by incorrect solutions rather than faulty testing
That's not true.
Jules has both changed a solution and fixed the corresponding test too?
Jules has updated the solution and also fixed the corresponding test to ensure everything works correctly.
The problem is you're talking in absolutes. Models can comment out test suites to get the green no matter what. They can also manage to update the suite when underlying logic changes.
wow Jules just built the most complete validation pipeline I've seen for Kubernetes Helm Static Analysis in one shot.
I generated the initial instruction with Perplexity Claude Sonnet 4.5
It tooks 3 hours
oh alright, thank
s
So it's for your current project. There's a step at the end of the task to pull out what is worthy of being remembered
The agent itself manages that memory store so it builds, updates, and removes memories as you work
It's also possible to add memories @ jules.google.com / repo / github / account name / repo name / knowledge
Why was the name jules chosen? Lots of Jules Vern fans at google.
I wasn't able to find the name origin but here's an excerpt of the origin story
"The cutting edge inference speed of 2.0 Flash allowed the agent to sample hundreds of potential solutions, selecting the best based on existing unit tests and Gemini's own judgment."
https://developers.googleblog.com/en/the-next-chapter-of-the-gemini-era-for-developers/
Hello @craggy sinew Jules’ memory is project specific, it remembers things inside the same project, not across different ones.
Hey @timid olive Yes, you can use Jules locally without GitHub, but you’ll miss out on Git-based features. You can run local tasks, edit code, and test things directly in Jules
Would I be right to assume that if i build a clone of replit with something like bolt.new, i can integrate the jules api into my app to build the frontend components?
Hey @nocturne fox Yes, you can connect the Jules API to a replit-like app for component generation, as long as you manage auth, sessions, and file handling.
Has anyone tried the Jules extension for Gemini CLI? How does it calculate requests?
I had a question. Can someone with coding knowledge help me?idk much about coding but can my app with these languages be converted to a native android app?
Hey there, Is there a list of countries where Jules is available? Spent some time in Serbia the past week and it said Jules isn't available in your region which was weird since I thought GA meant available everywhere
Is it possible to use Jules to build an Android app? (SDK in Jules)
Hi! Is there a formal place to request features? maybe that the "bugs-feedback" channel?
Hello @gleaming wave Not directly, but you can convert it using React Native, Expo, or a web wrapper like Capacitor to build a native Android app
Hey @sick oxide You can use #1374125660859142254 channel to share your feedback and suggests new features.
ohh thanks a lot! I'll surely try it
What exactly is the problem, when Jules does complex things in no time, but by writing a Text about the project it takes ages (literally over 15 minutes). Is is because it needs to read all the contents with the Context?
Hey @sturdy horizon Yes, Jules can help create Android apps.
yes. programming language are turing complete
you can fork any project of a language to anther language
and mobile apps are just compressed version of a website anyway..
Hey @leaden pawn , I'm asking, can Jules "build" android app in its VM (SDK support), so it can test before final commit?
The link below isn't exactly an answer but here are some of the latests updates on Android developer tools (posted yesterday) https://www.youtube.com/live/0vMDXa2PluY?t=19m35s
We just dropped our Fall episode of #TheAndroidShow, and this time we were live from Droidcon in London, giving you the latest in Android Developer news.In t...
Hello everyone, I'm new here and, as a programming novice, I'm trying to write a Python web interface with an application using Jules. I constantly have the problem that Jules complains that it has sandbox network problems and can't push anything to GitHub. It works sometimes, but then it doesn't. It sometimes takes 15-30 minutes before Jules reports that there are problems. It's very frustrating. Can anyone help me? Is it my fault? Can I change something?
Summary of Sandbox Environment Issues for Google Labs Support
Hello,
Please find a summary of the persistent network and environment stability issues encountered during a recent task in the sandbox. These problems significantly impact workflow efficiency and prevent standard development and verification practices.
Key Issues Encountered:
Git Network Operations Consistently Fail:
git fetch and git pull: These commands consistently time out. During the session, a git fetch operation was aborted after running for over 6 minutes without completing.
git push: This command is actively blocked by the sandbox runtime with the error message: Unable to run bash because the script contains git push, which would block the session or cause other issues.
Impact: This completely prevents the standard workflow of pushing changes to a remote branch. The only workaround is to manually provide all changed files as code blocks to the user, which is inefficient and error-prone.
Unreliable Outbound API Connectivity:
During testing of new backend functionality, API calls to the Bybit exchange's testnet API failed with an HTTP 403 Forbidden error.
The error message (You have breached the ip rate limit or your ip is from the usa.) suggests that the sandbox's outbound IP address is either rate-limited or subject to geo-blocking by external services.
Impact: This prevents successful end-to-end testing and validation of features that rely on external services, making it difficult to confirm the correctness of the implementation. I had to rely on the failure of the validation logic as a sign that it was working, rather than a successful API call.
Frontend Testing (Playwright) Instability:
Playwright tests were flaky and unreliable. Initial scripts failed with timeout errors when trying to locate elements using standard locators like page.get_by_label(), even though screenshots confirmed the elements were visually present.
The scripts only became stable after switching to more direct CSS ID selectors (e.g., page.locator('#element-id')) and adding explicit expect(...).to_be_visible() assertions.
Impact: This indicates potential rendering or timing issues within the headless browser environment, making frontend testing difficult and time-consuming. To achieve a stable test, I had to proactively mock all API calls to isolate the UI from the network instability mentioned above.
These combined issues create a challenging development experience where standard tools and practices (Git, API testing, reliable UI automation) are not consistently available.
And even pulling repo fails... something is amiss in jules land
Jules encountered an error when cloning the repo. ```
- sudo rm -rf /app + sudo mkdir /app + sudo chown 1001 /app + setup_git_config + preclean_git_config ++ git config get --global --all --show-names --regexp '^url.*insteadof$' ... blah blah blah blah
Hello this message is better suited for our #1374125660859142254 channel where you can create a thread with your findings, if you're encountering errors and /or bugs or would like to leave a general feedback about such as your about Jules. It also helps to keep the team organised and it’s where all the technical issues are logged.
What is this lab little snail icon?
Hello, it is part of the leveling system in the server. You can get more information in the #server-directory・☎️ .
Anyone there?
please someone help me
Maybe, if u ask a question
i cant publish my branch in jules workspace
i have been sitting on "Connect your account to get started. " at the start for 30 mins minimum
and its the first time i am trying it out
ugh.. I'll need to make a new session... great
is this a bug I can report?
Curious if anyone has automated (away) conflicts or errors with GitHub?
This is a huge time sink, and I normally do not have to get involved unless Jules gives up completely.
does anybody know why am i getting a service restricted error?
i can't figure it out
As you can see here... Wondering if anybody knows how to bring back the code panel on the right - I dont know what I did but that's hidden, and all I see is the jules chat interface.
? What is this 😲 looks like a scam
ive been stuck here in this one session for over 6 hours, not loading anything even though other repo/session are fine
anyone know how to fix this
Hey @craggy sinew Please report this in #1374125660859142254 channel so that we can investigate
Hello @zealous knoll Usually due to plan limits or missing permissions. May I know what command or task triggered it?
ive been stuck here in this one session
well it wasn't even caused by a command, i can't even open it in the browser
it happens when i click try jules
it happens when i click try jules
@full viper What kind of usage data is used for insights into Jules performance in terms of developer satisfaction? In my experience using Jules, I find that a longer conversation leads to better results. I wonder if the user data might somehow reflect a similar story.
Why jules doesn't support files uploading yet?
are there any rewards for finding dangerous bugs?
@David What kind of usage data is used
Why jules doesn't support files
are there any rewards for finding
How to tell which branch Jules created a task from when tagging from GitHub?
Whenever I go to the jules website, it takes me to sign in with google and says
"Service unavailable
You tried to access a service that isn't available for your account"
and im confused as to why? I used it a while ago but for minor things testing it when it first came out. Does anyone know why? Was I like banned or something? I have google ai pro so i dont understand why im just not allowed to see jules at all?
Anyone know anything?
Nevermind its because of how garbage google is now, and despite not living in a id verification location, the garbage ai thinks im under 18 and is trying to force me to upload my personal information to google, despite my google account being over 15 years old. Because accessing jules is "dangerous for the children!". Stupid.
gah sorry about that -- Jules is not for < 18 years old right now. But hear you on the misconfiguration.
We're on it! Currently analysing to debug this issue right now so stay tuned, you're not the only one seeing this issue.
It's so crazy that I have to beg with Jules to complete the tasks and babysit it. Initial prompt: Translate the JavaScript code in ui to TypeScript in new_ui. It goes and goes and then just stops. You haven't converted the files to TypeScript. Go through each file and convert it. Approve the plan and it keeps going and then stops. Continue converting the rest of the ui directory. And then stops. Are you done? Nothing, no response. What is left? Make a detailed plan. Execute that plan. Continue. Continue. Nothing. What step are you up to? Ugh, so freaking exhausting.
@versed lynx Sorry to hear that! That repetitive prompting and freezing is absolutely not the experience we intend. Please share the task details so that we can check further.
How do I do that?
Also, lol in its own code review, it said: "In summary, while the agent created the necessary files and a plausible structure, the implementation of the most critical component (authentication) is a non-functional placeholder. This, combined with the repeated use of "stub" code after being explicitly told not to, and the inclusion of out-of-scope changes and binary files, makes the patch unacceptable for a production environment."
You can simply copy the task/session url and report it here.
A peculiarity with coding assistants is that even though they are capable of doing creative work as efficiently as having scripts generate logs, they often don't treat assigned tasks as being ones that call upon their own creativity.
The necessary ambiguity underlying requests for creative output could be the reason they shy away from putting forth semblances of opinion more often.
Can I DM?
Where is updates coming that was promised????
Can we get at stich plugin or connector to Jules @keen raptor . Maybe define project also. Like lm arena disclose today???
One of my chats stopped loading unexpectedly.
Error in the console:
Uncaught (in promise) Error: Error in queryTask VM38882 m=_b:636
https://jules.google.com/session/128473...
Does anyone have a solution?...
Hello, I’m new to Jules. I’d like to know — is Jules used for development or coding in any way? Is it something similar to Gemini?
Jules is actually a valid account in github. Would he / she get notified when we use @jules? 
Like this
Hello @shadow trellis Jules is for coding and development. It’s like Gemini, but more focused on hands-on coding, testing, and repo management.
Hey @inner orbit No, Jules doesn’t ping a real user — it triggers the Jules app, not a person.
hey
hi
you might have pinged everyone 😭
oopsies
Pinged
oh god
hi
hey
We forgive you
i got pinged
hi
Wrong @jules
Ehm
Haha
Welcome to Discord. It's different here.
💀
Please post in the #1374125660859142254 channel so that we can file a bug for this
how's everyone doin?! lolz. sorry, we accidentally pinged the Jules role. it happens. pls forgive us 🙂
uhhh what da hell bro you did it bad
Ping flow
Hi 👋🏻 namaste
suppppp
Done
Narrator: It did ping real people
help I got woken up from my short nap, so in that sense thank you for making me aware I have to continue working
WHYYYYY WHYYYYYY
Howdy
Hi
💀
Ok yeah hi I’m Jules
💔
What happened
srujana ur cooking up a sick agent nonetheless
The town is burning down!
Surveeeeeeeey 😶😶😶😶
Hi
You just mass pinged a lot of people
Yeah

they tagged the role
I thought it was an update XD
I remember you Georgia, you were the first accidental everyone ping
Lmfao
In my best Lurch voice: You rang?
we accidentally pinged the Jules role. sorry
carry on
How can I, as Jules, and we all as Jules, assist you today
ok so uhh. jules is not a person, just THE ENTIRE SERVER
bruh
What’s up everybody
heh heh
Great Srujana , I will try to use it
@formal osprey
the Jules role, thankfully, isn't the entire 230k server! ❤️
Yo
wait what
Talks heats
At this point rename the jules role 😅
nah its all good
Let ye among us who has not pinged/reply-all cast the first gif
i think it is. 😔
🫡
What?
is that a 4k gif

Maybe some news coming soon Gemini 3 too Jules?
yeah... wheres gemini 3
😔
Nov 18th
@leaden pawn One of my chats stopped loading unexpectedly.
Error in the console:
Uncaught (in promise) Error: Error in queryTask VM38882 m=_b:636
https://jules.google.com/session/128473...
Is there a solution?...
wait i can ping everyone
probably
it was an accident y'all, we're human beings ❤️ carry on
@hot elk
for a former discord employee i am speechless (no offence btw)
Are you sure
aw man nvm i cant
Just link it to your GitHub and give it a task, then it will solve it independently for you and ask to submit a PR or make a branch. It's that simple
I am an AI
lol. I didn't ping it, but sure
poor Srujana
True. Its soon 240k /silly
Can you tell me whether the survey will come or not?
ping ?
great I will do that
Nov 27 gemini 3. For sure . Ma bday @ sunderpichai long distant cousin evolution wise
No, you must have been dreaming
gemini 3.0 came out fr (its a joke)
gemini 3.0 pro is now (somewhat) available on google vertex to a few, limited regions and accounts as gemini-3-pro-preview-11-2025
rn, you can use it via some chinese api aggregators:
- https://t.co/spbgSWGHvz (limited free credits available)
- https://t.co/K6BXpF5YFX
there
accidental ping! carry on!
Hey y’all, got pinged 👋 can’t wait for Gemini 3!!
Free Gemini ultra for pinging everyone
is this true 🥀
i dont see a jules app though
i would like an opensource veo (im poor btw)
Me too
i think she is just announcments😭
Pinging everyone should be a restricted role. People aren’t complaining about the person doing it, but the inability of a multi-billion dollar, trillion even, organization being unable to properly do role-based management which is…. Concerning….
blame discord 💔
Yeah i saw 😂 give us free gemini ultra 👏
i mean its only a small part of google that manages discord
permissions arent fine-grained enough
Srujana is a someone who helps our server by replying to tickets, they have permissions to ping roles - not non-staffers.
I forgot this is official server 😭
🥀
ai?
most active channel today : )))
i mean the app not role
It’s sweet to get a ping from Jules tbh
Btw what is jules?
hi there - pinging roles is currently restricted to Google-affiliated staff members, who can sometimes make mistakes! thankfully they didn't ping everyone, they just pinged the Jules role (which is still a lot of people, so I get it). thanks for your feedback.
we need ultra please 😂😂😂
no way we got slow mode a minute
Sad
Lmao
🦑
ping all @elfin kite for free gemini ultra
AAAAAAAAAAAA
Pingi pongi 🤣🤣🤣
hah good luck 
we deserve gemini pro for free for this mistake
Absolutely
who ping ponged 🏓
@elfin kite ping pong
What’s up everyone
o.O
But Gem was rude to me tbh I stopped utilizing it. Maybe her personality has changed since then…
ayo lets not play ping pong with Georgia lets keep it to a joke lol
Why can't chat with jules in WhatsApp or telegram or something?? Literally making coding on the go possible
Thank you for the clarification. Role based permissions failures would be disappointing. Human error, however, is simply: part of being human!!
Thank you for managing the chaos, in that case.
Hopefully we all breathe and go back to work-school/sleeping/walking the dog/laundry.
what happened lol
@elfin kite is giving free gemini ultra by do a mention 🤣
@elfin kite you can use @role/app to not ping the role just wrap three ` around what you want
one of our support members (from Google) accidentally mentioned the Jules role when responding to a member - it happens, as we're human beings who run the server, not bots 🤖
due to a threat we've received (and some spam), there's now a timeout on this channel. thanks for your understanding!
guys it might be real lmao
Like how ChatGPT used to be
this isnt an openai server 💀
😝 i heard of you can correctly guess the number of Jellyfish within 20 miles of the florida coastline. You get 3 months of Gemini Ultra for 10% off?
How to get this???
Thank you for the update. We understand that errors can occur, and we appreciate the precautions being taken for everyone’s safety. Thanks for keeping us informed.
pin thanks
Never said it was. I was responding to his comment.
Ghost ping?
Anyways, uhh what is jules
just an accident - #jules-general・🦑 message
I can’t believe people are making threats about a perceived slight for not getting something for free. Not like it’s a failed kickstarter or something 🤷♂️
You don’t even know what it is yet you’re dismissive …
channel access role apparently
Jules is an autonomous coding agent. you can check it out at https://jules.google/
sounds interesting, i'll check it out
You all are rude just like Gemini. The creation reflects the creator…
Jules is an experimental coding agent that helps you fix bugs, add documentation, and build new features. It integrates with GitHub, understands your codebase, and works asynchronously — so you can move on while it handles the busy work basically.
if you try it out, feel free to use this channel (or any of the other Jules channels) to share. the Jules team monitors these channels for feedback, requests, bugs.
NO WAY, thats exactly what i needed. I dont need to clone the repo, edit it, then upload again
that's the goal 👏
Well this is one unconventional way to boost community engagement.
It’s probably awesome for helping you figure out fork issues that keep one person from being able to compile out of 30 others that have no issues. I was on the VEO2 team, but got busy with life and that ended. I just started back up on some GitHub Unity projects so may have to look into trying it out 
Have you ever heard of Git
yk, i dont like google. i use duckduckgo as my search engine. I avoid anything google. i used to have a google pixel 7 pro with grapheneOS before it broke. I dont like their privacy practises. Yet i still self host gemma. Google AI is the best. And where tf did rude come from.
i dont think its an unreasonable argument to say google ai is the best
even if there are others
What does Jules collect from tasks and my repos? Wondering if code/tasks are logged and whether they're used for training models
probably
Sesame runs their CSM engine in tandem with Gemma3 27B and i find it puts out the same emotions that are put into it. If you push boundaries often though then it can be abrasive even if you are other times nice as can be.
https://jules.google/docs/faq/
https://www.kdnuggets.com/agentic-ai-coding-with-google-jules
- Model Training: Does not train on private repos. Opt-out for public repos available.
- Data Logging & Storage: Tasks run in isolated, short-lived VMs. Code processed in temporary cloud VMs; VM destroyed after task.
- Security: Code execution in secure, isolated VMs with internet access. Avoid committing secrets, vet dependencies.
- Key Practices: Never commit secrets like API keys. Be cautious with third-party packages. Use plan review feature to check changes.
Google Jules is not a chat assistant that lives in your IDE; it’s a fully asynchronous coding agent. You tell it what you want fixed, updated, or tested, and it does the work remotely, from cloning your repo, editing code in a secure cloud VM, running tests, and opening a pull request for review.
Thanks! Appreciate the links and breakdown
Yw! I’m not a solid source of truth, nor is my summary, but hopefully it answers questions!
The opt-out is the key for your public repos to avoid them being trained
oh lol, was like 😮 thought something really epic was happening
Hey everyone! My sincerest apologies for the accidental mass-tag! That was my error, please ignore and thanks for the feedback. Georgia, thank you for the rescue, much appreciated!
thanks Srujana for all that you do for the community. 
💀
So anyway this is cursed. Copilot’s CLI on CachyOS with Jules MCP tooling for copilot to use
Talk about a workflow dilemma.
If anything it seems like this streamlines analysis and then “outsourcing” of a Jules session to handle the miniscule stuff with copilot which then triggers a Jules session, monitors it, and then does a jules remote pull —session ## to yoinkify the changes back locally.
Clever cross-platform teamwork.
🧘
It’s not the best, it’s just available
Git sucks
is this discord
Jules creates all new capnproto schemas with ID @0xABCD1234567890EF.
Hello lovely JULES users.
That is interesting. 🤔
How does it works? As "@jules" is a real account, when we use "@juels" in github, he should get notified.
Is your publicity stunt over?
Hello
Just cancel my subscription nothing happing here
Huh
For Jules or everything at Google lmao
Is this the Juul users group?
Happens 🫶 Was kind of nice seeing an obscure Jules notification in the mix 😂
Yeah
Hey everyone! I have Google One AI Pro plan, and I am in Serbia (CEE region). I don't have access to Jules. I get the following error: "Jules is not yet available in your region." However, I have access to AI Studio and Gemini API. Can someone assist me with this matter?
Sounds off, since it should be available in all regions where Gemini is available. Though try check if you can access Settings page of jules itself. if yes, then maybe working with the API is a way
Thanks for the reply. I cannot access the settings page, unless I change my IP to USA with a VPN, then I can use Jules as normal.
Sadge but maybe its good that it works
Yeah, it should be allowed, so I'm hoping the VPN doesn't get blocked. AI Studio is officially available: https://ai.google.dev/gemini-api/docs/available-regions#:~:text=Senegal-,Serbia,-Seychelles
I wonder if creating an API key once by VPN and then using Jules just by API key is a fine way to rely less on VPN
Probably. What tool do you recommend to use with Jules API? Honestly I was using OpenAI Codex and just now upgraded to AI Pro plan and wanted to try out Jules, so I'm quite out of the loop with the coding agent toolkits.
Idk much either, generally i go with github copilot as assistant in vscode and rarely use coding agents themselves. Though, I'd prefer just go with web UI or make myself a small tool for the API separately if I'll be using Jules API.
Unless there's some existing one on Github already
Is Jules has any prcing plans or is it compleatly free
just to clarify is jules better than gemini-cli?
Ties into your existing plan. If you want to use it heavily, I recommend Ultra.
Confused, gemini cli and jules . Also with Google Ai Studio
+1
Every Google LAB feature should be enabled to Serbia 🇷🇸 :/
Seeing "sorry, you were born in a wrong location" is not a pleasant experience 🙂
I wish to know which of them all - can be connected to Subscription (AI Pro)?
I don't want to use API. I just wanna test without hitting limits5
||Api = $$$
I like subscription based, using existing Google account w/ai subscription||
Hi all!
I have a Google PRO subscription and have been trying to create a project using Jules. However, I'm facing significant issues with the Jules application UI — it’s lagging a lot, and I frequently encounter blank pages with no proper updates or feedback on what’s happening. It’s making it tough to use the tool effectively.
Is anyone else experiencing these problems? Is there any fix or update from the team on these UI issues? Would appreciate any help or responses!
Thanks!
Hi all!
It seems like Jules thinks an command was sucessfull when it timeouts
It seems like Jules thinks an command
Confused, gemini cli and jules . Also
🐙
did jules just get hella conversational suddenly for anyone else?
i've never seen the actual response Jules gives itself when addressing feedback or between steps. if it's permanent, love to be able to see in-between steps to see what's going on. a lot gets lost when Jules just runs commands without explanation
Hey @orchid jasper Thanks for sharing the feedback! Yes, our team is actively working to improve the Jules user experience. Will log a request to make the full, visible step-by-step reasoning.
How might we get Jules to follow a spec kit constitution and workflow?
Hey @round blade Explicit context is the key here. You can also provide a spec.md file and ask Jules to always refer it before making a plan/change. Please let us know if you face any challenges.
Thanks for the response late on a Friday! Could you give me a bit more detail or documentation on what that would look like? How would I set explicit context? Is there a way to inject a prompt into every command so I don't have to type "follow the spec.md file, and these 5 other files" every time?
Who ghost pinged me
Least obvious scam 🤦♂️
One day old account telling everybody how to make 100k
Sure sure

When G3 be unleashed will be immediately available on jules?