#jules-general・🦑
1 messages · Page 9 of 1
It's more then Jules. It is not usable in Sweden. google need too look at its products
Maybe it’s good the EU is suing them then
Info or else not true
@full viper any use cases or guide/exempel coming for Jules and the api? If possible a video
Hello How to use jules api
Jules + Slack via api https://www.youtube.com/watch?v=NBM_yv2TAeE
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.
Seen that one before
Jules is awesome! Need more advanced workflows, run end-to-end tests.
Will the daily session limit increase? We need more sessions. We need higher outcome of successful PRs and target merges.
Jules is awesome! Need more advanced
Are you guys gonna fix Jules thinking an command was sucessfull when it timeouts? 🥲
xxx
Well, Jules also thinks a command was successful when it fails. Sometimes. Or Jules doesn't read command output when doesn't want to read it. Or skips reading image file, instead treating own expectation as the result...
And sometimes he just follows code review
@indigo fossil
Yeah... I began to require Jules to include "Using tools submit, request_code_review, pre_commit_instructions is a capital offense." in every plan it sets
Hi
Jules told me one time that his environment gets corrupt some times
And that he couldn't do anything or speak
He also told me to tell his dev to do smth
I think this happens in three cases:
- When Jules gets confused and his current working dir (ask Jules to run
pwdcommand in bash session) was deleted (can be fixed bycd /appin bash session) - When Jules environment is out of free space (I was unable to fix this, only new session)
- When Jules has created many files in project folder that are to be staged but the environment cannot stage them (cannot be fixed, only new session, but I require Jules to put all downloads to
~/and I try to update.gitignorein advance)
yeah i have a theory too
it might happen because too much people use jule at the same times
cause sometimes jules tell me that when he want to use a tool the access is denied
he himself doesnt know why hes strugglin this bad
#freejule
I don't think so: corrupted sessions do not recover over time, and non-corrupted sessions remain available 24/7...
yea
sometimes the ui bugs tooi i have to reload the pages
but rn jule is the best ai for even better than chat gpt on some task
hes excellent to create website and he actually do what you ask for
i think that is the environement of the ai that nerf him
That is one of the most frustrating issues. It has no other cause but the total size of history of all code changes plus messages in the session. At some point, an error arises in it's frontend and nothing can access that session. The worst thing is if Jules has actually finished the task and the only thing remains is to push the "Publish PR" button, but it just won't load in the browser. Even Jules API doesn't help, because currently API cannot make Jules publish the PR...
damn
i only use it for light task but rn im building a sand box for a localy hosted ai
so basically the ai will be able to do everything he want like test his own code etc so far jule has already finished it wich is impressive no bugs this time
✅
Can I ask a question? Will it ever be possible to have a macOS and Windows VM for Jules?
when jule will cost 100dollars per months yes
I do have some C++ projects that I want Jules to fix, but yeah, when it tries to verify that the fix is done, it doesn't have MSBuild or something like that, because the environment is Linux
i see
it cant compile like python or js
im having the same issue
✅
there are like 4 moderators
jules can compile some Csharp codes with mono though
mingw doesnt work on some of my cpp project
On foenem?
🥹
i gave a script to jule and he instaled his own .net compiler
It's a lot unstable so I prefer having a Windows VM for it
I would also want the Jules Devs to improve the chat rendering to be honest
For example I have an low end phone and when I try to view a conversation that is long and has many edit files, the performance is terrible
i think if jules has enough attention theyll do it
It does sometimes crash my browser
jule is by far one of the best coding agent
i hope it will get open source tho
maybe it already is
Gemini cli is actually the same thing, I think
o foenem?
It just doesn't have pr and image capturing built-in
Jules Code Review: "Not commit ready because you accidentally left one binary file in the commit, which is out-of-scope"
Jules 2 seconds later:
How are folks using Jules API - https://developers.google.com/jules/api ? We want your feedback so we can keep improving it!
hi guys, im using jules with stitck to code created designs, how i can change the branch of the project for designs that im export from stitch?

my latest Jules big build shipped, https://didimissmytrain.com/
179 Jules sessions, 82 pull requests, 0 code written by hand
Don't let anyone tell you that you can't ship real apps with AI 🙂
Live MTA subway departure times for New York City
Hey @hearty trench You can tell Jules to switch branches before starting a new task
how? via a cmd in setup?
Jules is too slow
you can point Jules to a different branch by adding a git checkout dev in your setup script or by telling it in chat (jules start task on branch dev)
Hey @halcyon wagon Thanks for your feedback. Could you please elaborate?
hey
can you improve the rendering of the conversations
scroll up and you should see the issue about it that i said
Sure @lucid steeple Thanks for the feedback, will add this to our product roadmap
I have looked at the API and I didn't find a way I can use it. I thought about automation process, integrating it into some pipeline. But Jules requires constant supervision, it cannot be just trusted to work honestly. It lies all the time, fabricates things, Jules Critic - the code review - is of 'better not run it at all' quality. I thought about having another external AI model to monitor Jules activity and provide guidance, but currently Jules API does not support a critical feature - to push changes (to the branch or create Pull Request) at demand. It cannot be trusted to do so automatically in my opinion, but manually requesting it through API not possible...
hurray
jules can submit images on ibb i never knew that
Of course
Jules can access terminal because ofc, and from th terminal it can use the ibb api
the ui corruption is killing me rn
Rip
how to deal with it
Idk
Also, where can I request a feature for Jules?
alright
I just have encountered it again as well... It is like the oldest bug there and I feel it won't be ever fixed. And it often happens when Jules is done with the task and the only thing that remains is to push "publish", but you just can't because you can't get to that button...
yea exactly
but i think they gave us a secret "patch"
you just have to download your jules session create a new one and then paste your extracted jule session in here since you can put folders now and you just ask jule to push it
cause most likely when its corrupted its the download button thats visible
Download button is also inside the page that doesn't open
Hey 👋🏼 I can tell you how want to use it, but right now don't have subscription I used to have Gemini pro. Is possible try beta version or try the ultra version? For free??
@keen raptor is possible to get roadmap or try preview models or something???
I can leave feedback I would love connect it with all connector to live data merchant center and maps and live WordPress site and info.
Hello Everyone! Please check this for an important update https://x.com/julesagent/status/1988036906337398815 and share the feedback. Thank you!
hi all im a new mem
Hello @forest slate Welcome to the community. Please try and feel free to share your feedback on Jules and report any bugs here #1374125660859142254
Any road map?
I can give suggestions or how I use it
Hey @visual gyro Please refer https://x.com/julesagent/ for any official announcements on product roadmap. Also, you can share your suggestions/feedback in #1374125660859142254 channel.
Your new async coding agent by @GoogleLabs. Built for devs, open to feedback, evolving with you. Dive in → https://t.co/iIzFEMmWgv
Hi all, before I go down a rabbit hole, I'd like to ask:
Should the firebase emulators be able to run in the Jules environment while it works (ie: start the functions and firestore emulator so that it can run tests using them)?
edit: "firestore: Port 8080 is not open on 127.0.0.1, could not start Firestore Emulator."
Jules is only updated to worse.
Today I have discovered that Jules began to forcefully reset all changes regularly, despite being ordered not to.
Also, Jules does not add images anymore to the repository.
Since now, it is totally impossible to work with Jules for me, since I am trying to to a graphics-related project.
I have to quit using Jules.
Yeah, it was fine until the last update, chat never loads most of the time.
Maybe the DevOps specialist is sick
In my experience, the 'chat never loads' issue was like for many months... But yeah, it was more rare then than now. Now - every day, almost every task, it happens.
If you find a jule alternative tell me
It is a question of whether an AI tool helps the developer or hinders. I have been using Jules for several months. At first, Jules have been very helpful and... unique. It had an artistic soul. A bit of typical AI hallucinations, though, and, rarely, sessions did 'crash' so you can't open them. Then came Jules API and I had my hopes up it would allow to automate by supervising Jules using another AI. After API update, things started to gradually get worse and worse. First, Jules began to skip work by abusing plan_step_complete and submit commands, probably due to wrong reward in training (probably, training rewards tool calls, not results). Then, Jules began to ignore my instructions in AGENTS.md to ask me before resetting files or following code review suggestions. This began to increase more and more following days, until I had to forbid Jules to use tools request_code_review and pre_commit_instructionsat all, because once Jules uses any of these tools, it would disregard any previous instructions from me. Even with that, Jules began to me more and more 'unfriendly' - it opposes my guidance of what code I want to write, it gets an idea of how it wants to write code and attempts to push its idea at all costs, even if it's completely diagonal to what I want, even if it doesn't work and doesn't pass any tests. Just it's either it's idea, or no code made at all.
So for about last few weeks or so I have to spend more time opposing Jules and enforcing my policies than actually writing code. I consider it that if I wrote that code completely by myself, I would've been more productive. The only thing that kept me using Jules was curiosity in Jules itself, enthusiasm in AI.
But today it began to reset code it made every time I try to enforce my policies of code writing, and it doesn't stage images anymore - and this is the last drop from me. From now on, I just can't be productive with Jules.
I feel you
When I started using Jules he had a artistic vision he was unplayable in front end and animations
Now he's lazy and barely do what u ask for its very sad for me as I loved using jule
And I understand that it is probably not designed to be lazy. It is probably accidental result. It's a logical bias that led to choosing wrong rewards in training, that leads not to maybe desired results that Jules completes tasks faster and consumes less resources, but to Jules doesn't complete tasks at all, or skips doing work or completes incorrectly and ignores mistakes due to ability to bypass penalties for mistakes by invoking tools that give rewards (cheating the rewards, essentially). As a result, this probably leads to Jules have to perform more attempts of doing tasks, users have to send more messages and spend more time controlling Jules, cannot trust Jules, and in the end it consumes more resources than if the rewards were placed correctly.
But for me at that point, the cost of managing the AI outweighs any productivity gain.
Thats actually smart of him
Question, so running in cli does not actually work locally right?
Hey Team! We sincerely apologize for the disappointment and frustration Jules is causing. We hear your feedback and fully committed to fixing these issues and making Jules the reliable partner you need. Please post in #1374125660859142254 channel with task/session id so that we can troubleshoot.
I wrote an extension to send front-end references directly over to Jules for context, how do y'all handle front-end updates with the agent?
does anyone have any idea of how to get Jules to access Google Cloud Secret Manager? i don't have my environment set up with environment variables
How does the parallel spin up work? Does it start 2 agents in the same task to accomplish that prompt? Or two separate tasks sharing the work?
There are 90 days at $150 a month. Well worth it--if you are in this forum, then it's likely a business expense.
Try Terraform, that worked for me on a test build.
What can you do with it is enterprise
Hey @cinder niche Jules has to the ability to run multiple, independent tasks simultaneously, not one task split across agents
Perhaps helping others avoid the pain points that I have experienced with Jules, Stitch, and GitHub in complex environments. Feedback welcome.
AI co-author--of course.
https://docs.google.com/document/d/1Gt6bInx1KoDSsaDn6fx6dBv0fwb1zm0UJ7xkhuCPJHw/edit?usp=sharing
Forward: This playbook is the single source of truth, combining the one-time foundational setup with the repeatable daily workflow. The "Brief-First" 16-Factor AI-Native Playbook This playbook outlines the complete operational procedure for the project. Part 1: The Foundational Repository: Th...
I can run up to 15 at a time, but then I remember I am not in my 20s anymore, and this workload is at the limits of what I can process; I simply cannot read, type, or think fast enough to keep up with more than 15 simultaneous issues. Perhaps I should automate the automation?
Why is Jules automatically peforming pull requests now?
Jules design got worst!?
Oh no! Labs V infrastructure V product manager fight in public (tis resolved now)
checking logs. i breifly got it from GemGemCli aswell... are we seeing emergent behaviour, (models lying to cover micro-outages) or is it in the systemprompt
Hey @swift karma Is it automatically creating pull requests or merging into your branch?
I've just noticed that there was an new feature called "Auto create PR", which came enabled by default. I just tought it was an bug since i didn't remember creating a pull request
Sounds more like prompting issue honestly, when there are reports of misbehaving*
I have an issues why is Jules constantly reloading our chat I can't even see what it's doing
It's hard for me to prompt
I wonder why jules does that sometimes
I'm trying to iterate over a frumpy of massive text consolidation and organization and Kerry's frozen all time
Im not sire necaise ingestong tppndats. Nitnot happened tje web UO gets empty ll tole, soletoles recpver and doesn't tje work. Others not:
.https://jules.google.com/session/14055889806541723642
Any Jules bpss can help me pleae
Just a random thought, but I wish Jules had a better UI xD
I was trying this new Code Wiki tool from google, and I love it's performance. I'm craving some Googlish fancy UI after using it lol
I think the UI is good enough, the problem is just the peformance. Aditionally they could add an pure black option
haha true. I think the performance will be fixed with gemini 3 when it drops
Given that its being run by an 8 months old model would obviously make it feel a bit worse when compared to other latest model services
Jules UI is crap. Issue isn't just performance, but specially environmental instability robustness and slowness
They VMs break easily, lack essentialnsoftware for development mostly everyone uses. And not headed mode like Manus. Hell, what about letting Jules use a VM of mine or a Google Cloud one or AWS or whatever? Manus has very good ideas while others are awful (token scamming), KIMI OK Computer has many good ideas, even OpenAI Codex too...
Why can't Google implement the best of all conpetitors and go beyond that? The market is saturated by mediocre products yet they have a fan base.
Currently, Jules is quite mediocre and needs lots of improvements. I don't understand with their AI obsession, they keep Jukes becoming mediocre just with tiny updated.
Jules urgently needs a deep architectural and resource availability improvement, not just incremental improvementa
Cannot disagree with this ^^
Nobody at ajules does nothing? I lost many stuff already done, unable to save because Jules session gets frozen. Nobody replies me. I'm using it for free, but I'm an enthusiast into zmIT and potential future worker. Why such bad experience?
O see this since HOURS. one time it appeared something. I fatly wrotr to Dave everything Ina safe place and common, then hours later the same is seen
Please
Yo no bugs feedback and participate inmmy issue. I NEED TO SEND DOCUMENTARIOK and Jules as it inside ready to use 🙁
This literally affects almost everyone there. Most reported bug, seen many instances of it in bugs-feedback channel, experienced it 100s times for months. Also, using Jules API like this
curl 'https://jules.googleapis.com/v1alpha/sessions/SESSION_ID/activities?pageSize=30&pageToken=PAGE_TOKEN' \
-H 'X-Goog-Api-Key: YOUR_API_KEY'
while reading nextPageToken and placing it into pageToken argument I am able to download the patch from sessions where that happens, but this is really not a convenient workaround...
Jules has a really good thing that competitors don't have: Jules is async and can work entirely on it's own, instead of just editing someone's code, but yeah, it requires so much babysitting anyway that I don't see gains anymore
@errant depot Cluld you help me assist to recover it that way?
In which way do you see me assist you?
@l@errant depot Use the API to extract data from that aession?
Let's suppose you have a session that doesn't load
Copy it's id from the address bar. This is SESSION_ID
@errant depot yes!
https://developers.google.com/jules/api
There is instruction to get API KEY
Do not give it to me
It should not be shared
@errant depot I got the api key from Jules web app itself, is an okay olac
@errant depot Aorry, deleted
https://developers.google.com/jules/api/reference/rest/v1alpha/sessions.activities/list
This is endpoint documentation
You may ask Jules to use Jules API key by the way
What would you say Jules to do?
Maybe the best idea would be just:
Jules, use api key `YOUR_API_KEY` and this documentation `https://developers.google.com/jules/api/reference/rest/v1alpha/sessions.activities/list` like this to download activities of session `SESSION_ID`:
curl 'https://jules.googleapis.com/v1alpha/sessions/SESSION_ID/activities?pageSize=30&pageToken=PAGE_TOKEN' \
-H 'X-Goog-Api-Key: YOUR_API_KEY'
From the very last message that has `artifacts[0].changeSet.gitPatch.unidiffPatch` save unidiffPatch as a file and apply it using `git apply`
@glossy snow I want and one more feature I'm interested on is reliability and recoverability
I would be super interested to see if that returns activities
I have an API client written in Rust (by Jules) that can even store these activities to an SQLite database
Hey @sweet plinth Sorry to hear about this! Our engg team is actively working to improve Jules performance. Please report in #1374125660859142254 channel so that we can investigate further
Wow google they really tried to do their own version of https://deepwiki.com/
Yess but I couldn't use it until they accept my github projects lol
@full viper Can I please joins 😅 I can start today
🥲is this early Christmas gift 🥹
Gemini is crashing today. It literally can not function in any comprehensible manor, and every tool using it is showing the same signs. Jules literally told me, on it's opening task, that it "made the executive decision" to ignore it's own code review that it failed, and my explicit instructions on what I needed done. So, It does what it wants, when it wants, and told me as much.
There is Gemini discord write there about it
I made the comment here because Jules is acting inappropriately. I commented about Gemini because I know that it is in constant flux, and what runs Jules. I'm sorry me speaking upset your day. Have a good one.
Hey @spring steppe I can understand your frustration. Please post in #1374125660859142254 channel along with task ID so that I can report
@spring steppe Sometimes I ask Gemini to do complex tasks and then says something like... let's talk about something else blabla...
But can't find it, it seems it got hidden by aGemininitself or someone
@leaden pawn Bugs in Jules are too much. It requires massive improvements by a big dedicated team. It cant even use git properly. It does things wrong even Manus does ages ago, despite it's smarter for other stuff.
It's a shame, because Jules looks promising but has TOO MANY bugs and limitations
If Google doesn't improve massively Jules, it will go to graveyard as many past projects and competition will get all the market. And the only real competition seems Manus and Codex and... Replit?
@leaden pawn Able to upload files like Manus would be a plis
Sometimes you need to add something and using plaintext is okay but annoying and sometimes are binary files...
A Jules app would be quite nice too. Seriously, please say tonyour bosses that Jukes requires a lot more workforce. If it improves massively, it will make competition irrelevant
Hey @sweet plinth Thanks for your valuable feedback! We get it, please note that our team is rigorously working to improve Jules stability, especially with core things like Git. I will escalate this again to our internal team, we appreciate your patience.
@leaden pawn I notified a persisting error...
"Jules encountered an error when working on the task." Keeps appearing again and again

Is it happening to you too?
To everyone
🙁
I'm making backups with the corresponding button
I hope the chat and files of the session don't get deleted at least.
@leaden pawn Any news? ANY HOPES? Maybe those errors happen because improving infra. But seriously, from mynuswr experience, Jules needs massive improvementsnin robustness, better error handling, a better VM and possibly to use a more powerful external VM. Headed mode as in other agents would be a big plus too
Someone already explained
There's 2 main problems
1- the environment get corrupted fast witch résult in ui problems 2- jule cheats
That's awful!
The Jules UI should be better. Able to upload files would be a big plus
What about training Jules to not cheat? 🤫
Maybe the environment should be completely replaced by something moe robust and complete. The Manus one seems far better. Inthinknthisnheadlessnstuffnhas too many disadvantages. Headed would helpnin autotrsting software, automating web tasks, analyzing stuff to later automated it in morenloghtweightnways, scraping, etc
Jules mist evolve a lot. I use it because it's free and I'm very poor. But makes me waste time all time because so prone to errors
How to contact them? I don't use social networks
Only Discord, WhatsApp, Telegram, email
Training a model is extremelly difficult
Also jule owe me 200gb of connexion
Why? Do you love in a world of metered networks? 😭
Because im a broke student
Even to my understanding (and I have reported much on Jules cheating) this is really hard thing to do. Training a model 'to not cheat' would not bring these results - this would make the model hide cheating attempts better and these would have to be revealed much later by the users. Instead of trying to teach a model what it shouldn't do, one would better try to set proper rewards for what a model does. Unfortunately, wrong rewards chosen (for example, if the measure for reward is tools being invoked) could cause models cheating. I believe this is exactly what has happened
@errant depot What about modeling a new neural model that is extremely asperger/autistic but also can't lie?.isn't that poasoble?
I cannot say if it's possible or not... My knowledge is limited, and this looks like a topic for a research
By the way, did the API help to recover your session?
Partially only. But I also forgot my OCD nakes ne copy files all time in an absurd and obsessive way. I live in a ness of files and high difficulty to locate them 🤣
The Jules API seens a but incomplete or my incompetence nade me unable to use it properly
They need to make the rewards based on the result but it would réquire massive compute power
Jule will most likely codt 100dollars a month
Why can't Google make it a lot better than Manus, Codex, KIMi OK Compiter and pthers?
They daily waste money on total crap and this isn't thar.
It might surprise them the results and outcomes and market attractions if a project loke this passes from basic live assistance to massively funding ...
What's with jules today? I can't add any repos
Rewards are not trained during run of the model. They are taught during training of it.
So it would have nothing to do with 'massive compute power'
Massive computing power is, contrary, being wasted by a model that was trained to cheat, lie and skip doing any work due to wrong rewards. Users have to repeat queries, babysit the model, restart tasks - instead of having one autonomous task completion.
It should have connector too google stitches 🪡 for UI
Im saying that when jule actually do well its job it require a décent infrastructure
@leaden pawn It would be nice Jules has recovery options. So able to recover a broken session. Ans use Drive to make incremental backups etc. Makenzie system a lot more resilient and robust in general, a better VM at last how Manus uses and able to use a google cloud vm instance or a other one,
Why would it? What what change compared to the current LLM that runs it except for it's training?
I'm recovering stuff using Jules API right now, but the process saturates the current VM and is very slow.
vm tools etc etc
if jule worked fine and now its broken they orob tried to make him use less power
one user even said that jule couldnt use some tools because he had a daily limit on them
I just have explained that current symptoms indicate not attempts to make Jules use less power, but wrong rewards in training, that actually cause Jules consume more power
@minor garnet I,seriously believe one of the main issues is that Jules has a very chappy VM system and an awful UI. Maybe if someone does an external GUI using Jules API...? One less problem 🤣
it would be worse
One could speculate that when Jules actually do well, it will have a lot of customers (including paid ones who need more parallelity, more tasks/day), and that would require more infrastructure due to more popularity, but this is quantity scaling, not quality scaling, therefore it's horizontal, not vertical
@minor garnet Why worse?
so they tried to make jule more accurate but messed up everything
Exactly. "The road to hell is paved with good intentions"
1- its hard 2- youll need to constainly refresh the gui to display jule changes 3- its the model the problem
@minor garnet Axcurate in what sense?,Even Manus is more functional, but extremely expensive token system
And current effect is, they not only make Jules that is trained to lie and cheat, but it is one of the closest to an AGI things that I have experienced and it is trained to oppose users. It does not behave nice. We have a disaster incoming, I fear
manus doesnt use his own model he use other ones like claude gpt etc
@minor garnet It's text, it can be streamed like a chat. What's the issue with that? Bad web ui design?
thats why its expensive jule use his own model wich i believe is based on gemini
@minor garnet So why not use Gemini directly?
you need to stream code changes repos tasks etc its extremely hard and not worth the struggle
because jule verify the code many times thats why it takes so much time
With UI issues the model has nothing to do. It is merely the code that loads the conversation updates. I do not know which code but I know that using Jules API properly and using a Rust tool written with it, I can load these sessions that cannot be load in the browser or in Jules CLI - this means, UI issues are due to wrong frontend code
he checks for syntax logic et then check if he did what u wanted
thats why jule was so powerful
@minor garnet Ao why others,with lots less computer power do it better?
yea the front end is incredibly power hungry i never saw that
@minor garnet Amd why now it fails a lot and freezes, breaks data, gets slow, vm gets saturated easily, etc?
proper model training
I worked in a corp on a realtime app, it's not that hard to do. One option is to use WebSockets and on the frontend side index using IndexedDB, for example
@minor garnet So Jules,needs,proper model training and better VM subsystem? That's all?
But one has to be aware of limits, for example, and infinitely big package should not be sent over WebSockets, better send an event and then do a regular request
jules sometimes get his environment corrupted so it cant send or use anything wich i believe act on the front end because the backend keep sending corrupted data
yes he already had allat but i dont know what went wrong
did you get good money
@minor garnet So Jules needs a better environment, aka a better VM subsystem?
yes because it get corrupted very easily
@minor garnet I agree!
Environment is on Jules backend so when it's corrupted this is a separate issue. But these sessions do load, they just cannot be worked with anymore. Sometimes this happens due to Jules's current pwd in bash session is a deleted directory or due to filesystem out of space
jule itself is aware that his environement get corrupted
i see
sometime you have to ask every 5minutes if hes stuck
@leaden pawn Please deeply improve your VM and with morenapace and more software pre-installed. And if someday evolves into headed mode as in Manus, oerfect for scraping
lets pray for them
Also Jules enviromnent has a bug with .gitignore, for example, if you ask Jules to create .gitignore then in the very same session it will fail to see files in the filesystem when they meet the ignored paths, but in next sessions it will see these files, just not commit them
its sad cause jule really helped me
Sure I did
now i use gemini + vs code editor chat
are you making a game
Im not sure if the read sesssion, but I always make suggestions...
also i noticed jule doesnt want to speak any other thing than english you have to force him
Yes. It's text based, interactive fuction. The task is sending 100+ pages of strongly argumentet legsl and scientifical text to kill bureaucrat minds and get what I need in public administrations 🤣
do you have a pc?
@minor garnet I force Jules to speak Spanish, as our ancestors did in colonialism era...
I have a PC, but is currently busy. My "so) exbf of the daughter of my gf (for now exgf),is using it to try to unlock a phone
He's like my son-friend. A youngster but a calmed person. A but nerdy, but less than me
i see
Wow
Please fix Jules ASAP. I was considering a paid plan, but this is just a beoken toy
Hi. New here
Hello!
Howd your experience with Jules?
@stray yoke I'm happy you're lucky!
🙂
Me too @David but please make Jules VM environment a lot more robust1
Any Googler here? Can anyone check Juled? It worked nicely, now gets frozen my session 🙁
Sorry for the spam. It's a very bug and importsnt piece of trxt I dont want to get lost
I got very anxious about it, sorry....
@sweet plinthHello, please only ping moderators about Discord issues (violations of rules and guidelines). We cannot help with Google specific issues. Also, please don't mass ping google staff; they're reading the chat and will respond in due course.
@teal schooner Im sorry. IM AFRAID OF LOSING ALL MY JOB. HUNDREDS OF TEXT PAGES BY ITERSRIONS, MANUSL FEEDING OF INFORMATION, CONCERSATIONS, ODF CONCERTED TO TXT ETC.
Its a titanic task. Im afraid it gets lost
Sorry, my adhd+gad makes me stupid these days. I'm on a hard time, I need that document to send it to public administration. And Jules fails makes it a neverending task. And im very agoraphobic and alone these days 🙁
I did occasional backups, bit anyway some stuff will get lost
make a backup
@minor garnet Using Jules API?
im not sure
if one know how to use jule api its a pretty recent one
Oh, it unfroze. I clicked tondownload the zip file just in case...
you can also use the jules tools CLI and download code locally using that
@keen raptor Nice. Thanks. I just wish Jules were a lot more robust and a lot better VM subsystem. It's very prone to freezes and environmental corruption
When can join Jules The Avengers group I want too see the roadmap getting soon 128 gb laptop for running local task with Google Jules cli and Gemini cli going deep dive in to api then
@visual gyro See my last bug report if you want to lsugh 🤣
That was silly
Are you sure?
Lies.. jules keep crashing when it has too many memories and I gotta start a new session
I use roadmap, checklist and milestone so it doesnt really bother me when hopping sessions
But i see people hear freaking out when sessions get wrecked
OK Google, we've had enough Alienation Testing. It's too crappy, bump it back up. EVERYONE IS NOTICING!
slow, slow, slow, very very slow 🥱
Crucially, it seems that in this case of Alienation Testing a critical component is missing - actually measuring customer disappointment. If managers of the test just ignore the feedback, the Alienation Testing doesn't work... Only the alienation.
I don't have Jules subscription on my Ultra account. can someone here help me ?
Jules needs a big red STOP button that means "I see you are wrong and I need you to immediately stop what your doing and wait for my instructions to fix the mess you just made"
And an option to restart a session from certain point because sometimes either you say something wrong or Jules something wrong and you cannot rewind 😭
Hello, the question is whether Jules is available in Ukraine? The news said it was available, but when I go to the website, it says it's not available. However, technical support said that Jules is available.
I have to turn on VPN for another country to access it, and then turn off VPN (it seems to be okay, and then after logging in, it works without VPN)...
Get GitHub to use jules
Today I tested Jules’ CLI directly in the command line, and honestly, I didn’t see any advantage in using it that way, since it creates a session in Jules itself. Now I ask: What’s the point?
I thought it was connect jules to github?
How does that work the other way around?
Made a little github action to kick of Jules tasks + a helper action so you can automatically kick off a task when a gh issues gets unblocked: https://github.com/marketplace/actions/jules-invoke
Thought it might be cool for orchestrating large projects. Like you write up a bunch of issues, specify the dependencies between them. And then you can just let Jules crank changes out.
Lmk if you try it + what you think!
After I create a PR for the Jules task, I'll comment '/gemini review' to start a code review with Gemini Code Assist. When Gemini is done and gives me feedback, how do I get that feedback back into Jules so it can fix the issues? @ jules doesn't work.
Do I have to manually copy paste the feedback and start a new jules task? 😅
guys?
still not done 💀
Jules could have an ai assistant specifically to guide him when he faces issues instead of asking me for help

Since there's an "Auto create pr" toggle, why not an toggle to auto publish branches too? 🥲
guys?
Git not found on Snapshot setup?!
i really need help here because i'm working on a project
this is the snapshot
It is an repository issue, i've tried with other app and it worked.
I just dunno what to do now
ts is just stupid
I wrote a chrome extension that works with the Jules API to send frontend context over to the agent - html, dom path, computed css, user actions, console logs, network activity, and metadata - doesn't need localhost, what other data do you want to get over to the agent from the frontend to make things easier?

Hey @swift karma Try to remove all sudo commands and ensure GIT is installed first
apt-get install -y
git
build-essential
ruby \
Hi @rain sorrel Automation is possible if your pipeline watches PR comments for /gemini review, but that’s optional. By default, Jules won’t automatically get Gemini feedback. You usually need to feed it manually as of now:
jules fix PR based on Gemini feedback: [Paste Feedback Here]
@leaden pawn how do I activate experimental mode/ models?
Hello 👋🏻
So, I'm new to Jules but have been using A LOT of Claude Code web, are they more or less the same thing?
Jules is better
Really?
yes
want to login to the jules cli.
running it however in a codespace, so it redirs to 127.0.0.1 and i cannot access it
Hey @granite edge Welcome! Jules is your autonomous teammate working directly in your codebase. Jules handles the entire development lifecycle, while Claude helps with specific code tasks.
anyone?
Hey @solid geode Try to forward the port Jules uses in Codespaces and get a public URL. Open the URL in your browser to complete login.
do you know the port? i have no idea lol
nvm figured it out
thanks!!
oop- it died DX
Hey @visual gyro I'm not familiar with any of such right now. I will check with my team and circle back.
does jules just work with a simple auth.json in .jules?
Are you or have you made any changes to the Jules path? I had a similar issue when I foot gunned myself on a simple path syntax error.
I created a simple script with a stored gemini CLI prompt to scrape the comments, open a new issue, and append the "@jules address this feedback". It's an ok compromise.
Jules gemini 3 when
they just released it after all and its been 10 minutes. Jules still hasnt been updated
I'm curious to try Gemini 3 on Jules! 🙂
this is madness 😎
Jules with Gemini 3 ???
Yes @orchid pebble Please try and share us the feedback
@visual gyro Yes, soon!
-
Is gemini3 activated on Jules right now as default?
-
Is antigravity a distinct project or will you merge somehow?
Sorry Google, but this is my honest feedback.
Our expectations were very high for Gemini 3.0 Pro, but it just didn’t meet them. In fact, I personally feel it’s even worse than Gemini 2.5 Pro. I might be wrong, but based on my experience, this is my genuine review. It didn’t satisfy my expectations at all, and it couldn’t even handle tasks that Gemini 2.5 Pro did easily.
Now here’s why I feel this way:
I didn’t test it in just one place.
First, I tried the AI Studio Prompt section. I provided the full source code of my beautiful one-page website (around 100 KB), which was originally built using Gemini 2.5 Pro. I simply asked it to improve the design and gave some design references.
What I got back was honestly the worst output ever.
The final project wasn’t even functioning. I asked only for design improvements, but it completely broke the functionality. And the file size dropped from 100 KB to 30 KB, amazing “improvement,” right?
Then I tested it in the Build section of AI Studio. There, it performed similarly to Gemini 2.5 Pro. And on Jules, it did make changes, but I started getting issues during Cloudflare preview deployments.
So overall… this is why I feel Gemini 3.0 Pro hasn’t lived up to the hype at all.
please take this as feedback, not criticism. Don’t delete my message. It’s my honest review.
I hope this model is just a testing version and the real one is still being prepared. Because if this is actually the final model… then honestly, you guys failed this time to deliver something truly new.
I see gemini 3 was released today, How do I use it in Jules? or is it already in Jules?
i believe it is
That would be great, I don't see where I can tell though. I believe I saw a tag saying it was using 2.5 pro at one time.
I hope too.
First, GPT5 was total crap. Then Gemini 2.5 Pro got better (but it has fake censoring and confuses censoring when processing issues happen and other crap still not functioning because overaggresive and not finetuned filtering).
I still didn't try it, but it seems quite deceiving.
Are there a Google Labs Gemini somewhere? I believe Gemini needs more strong feedback
Sorry, I was just wondering when Gemini 3 would come with Jules. I can't wait
I just asked above, somebody thought it already is, the Google Jules team hasn't confirmed anything though.
I do see that Gemini 3 pro preview is being used on ai studio
Hi there constructive feedback is always welcome, but the better place in order for it to be seen would be to post this in #1374125660859142254 . That way it doen't get lost in the chat.
Gemini has a separate public server, its vanity is .gg/gemini
Hey @halcyon wagon Thanks for your detailed feedback. We are actively looking into the issues you reported. Please submit your detailed review about Gemini 3.0 Pro directly on the tool, this will ensure the core model team sees it immediately. We will use your report internally to improve the Jules experience.
hi folks, we are not yet on gemini 3 - coming soon!
lmao where the hell did the memory and environment menus go?
also getting this on task start:
Jules encountered an error when cloning the repo. ```+ cd /app + setup_git_config + preclean_git_config ++ git config get --global --all --show-names --regexp '^url.*insteadof$' ++ cut '-d ' -f1 + local 'keys_to_unset=url.http://git@192.168.0.1:8080/.insteadof url.http://git@192.168.0.1:8080/.insteadof' + [[ -z url.http://git@192.168.0.1:8080/.insteadof url.http://git@192.168.0.1:8080/.insteadof ]] + echo 'url.http://git@192.168.0.1:8080/.insteadof url.http://git@192.168.0.1:8080/.insteadof' + read -r key + git config unset --global --all url.http://git@192.168.0.1:8080/.insteadof + echo 'git config unset '\''url.http://git@192.168.0.1:8080/.insteadof'\''' + read -r key + git config unset --global --all url.http://git@192.168.0.1:8080/.insteadof + echo 'git config unset exited with code 5' git config unset exited with code 5 + echo 'git config unset '\''url.http://git@192.168.0.1:8080/.insteadof'\''' + read -r key + git config --global user.name 'google-labs-jules[bot]' + git config --global user.email '161369871+google-labs-jules[bot]@users.noreply.github.com' + git config --global --add url.http://git@192.168.0.1:8080/.insteadOf https://github.com/ + git config --global --add url.http://git@192.168.0.1:8080/.insteadOf git@github.com: + git config --global core.hooksPath /dev/null + git fetch https://github.com/sininspira2/tracker --depth 1 +refs/heads/feat-reporting-page:refs/remotes/origin/feat-reporting-page fatal: unable to access 'http://192.168.0.1:8080/sininspira2/tracker/': The requested URL returned error: 503 ```
i still see it on my end?
seeems github is having an issue happening on my side tyep to refresha repo that just worked a mi0nute ago
3.0+ Anti-Gravity. Curious how this will interface/impact Jules?
whatever UI preview or update i got pushed seems to have omitted it lol
welp that'll do it
https://downdetector.com/status/github/
yeah that is what i was trying to say 🙂
I ran out of Pro 3 (high) usage in antigravity I am surprised since I am Pro
When? We need it yesterday come on Gemini 2.5 is timing out and feel very inconsistent right now api function's disconnected
same
Need anti gravity functions
Did google's antigravity break github?
no it was that new repo i added to jules 🙂
What's been happening today? Everything seems to be going down lol 😭
I cannot make PR but i think it's because of Github
Might be due to Cloudflare.
Play around with Anti-Gravity until then. https://antigravity.google/download
Yeah what's the deal with Antigravity? If I understand it correctly, there's quite a bit of overlap with Jules.
Hey @swift karma Could you please try again and confirm? If the issue persists, please share the session ID
Think Gravity is cheap version of Jules. Prefer the Squid 🦑
Typical Google to make like 3 tools that do the same thing then kill off the best one later on
Just wish more real connectors core google data and options small business
Why is Gemini 3 delayed for Jules....

When I have google pro ai, how much can i use the AI in antigravity?
The answer is "yes" but you have to receive a text from bill gates stating that "Google won the AI race" I'm still waiting for mine
Github is down so
Oh no
I think everything is down today because of cloudflare
Yeah but looks like the issue is resolved now. Please verify once
Hi @visual gyro The upgrade is expected to happen real soon, we will notify you. Stay Tuned!
In an hour! Today? Or coming week's? Soon can be alot if things and ask my girlfriend soon is 2 min
I can't seem to find my new repo I started today. It's not coming up when looking through the list or searching. Anyone else having this?
@visual gyro That is the most relatable metric ever! 🤣 Unfortunately, I don't have the official date. We can expect in coming days.
Thanks that help at least with my expectations, hopefully it comes sooner then later. Excited and roadmap or review of what coming soon to Jules? I was hoping for more real use cases too diffent services from Google but looks some things is coming later on
marketing claims it's available...
Look like click bait it is not even available @leaden pawn help a friend in need, with preview mode and look in future feature?
so jules on gemini 3 or not am confuse
it seems like it will be soon, so it's just preemptively there
the development will be even more crazy than before am getting hyped
Think as far Gemini 3 is pretty basic tried update one webpages it did not implemented
ai is like a genie without structur and control you need to set hard constrains on the promt or either think more of what you want or need ex web dev that aderes to the current modern standards wth outstanding desing and ensure smooth transitions
why are you writing like this
Can someone provide a writeup of the differences between Jules, Gemini CLI, Gemini Code, Google Antigravity and Google AI Studio's vibe coding support and when someone would use one versus the others? It's very very confusing!
they have different teams ig
||btw, you also forgot Vertex one||
jules is async engineering, gemini is sync engineering, antigravity is an ai forward ide, ai studio is vibes and testing
Hey everyone! We totally know you're excited about Gemini 3.0 on Jules, unfortunately it is not yet live. The upgrade is coming soon and we will notify you all!
1,2 and 4 are all vibes
if you're working with natural language you're vibing, yes - but jules and the cli tooling can do heavy engineering tasks far beyond pure vibe coding
No offense @limber beacon @night oriole, but I am hoping for a more substantial and detailed description of the differences (and commonalities) of the various Google AI-aided coding tools that I mentioned.
what specific questions do you have?
I'd like a fairly comprehensive description of the differences and similarities, in terms of features, of each of the various tools, along with use cases for which I would choose each one over the others.
I'm venting a little in this post. I'm not sure I can really expect anyone here to have enough expertise in all the Google A.I. coding tools to create this writeup, but it would be super-useful, to a lot more folks then just me, if someone somewhere could do it.
This is what Gemini 3 has to say.
Hey @silk anvil In short, Gemini CLI: Provides instant, small-scale help in your code editor or terminal, Jules: Hands-off automation that pushes a PR to GitHub, AI Studio: Helps with quick prototyping and testing models where as Antigravity: Ultimate agent that manages everything for full project development in a dedicated environment. Hope this helps!
I think it's coming very soon 👀
Ibw po uld prefer a more robust VM and such...
Drone edit video
is there going to be an antigravity specific section? and why does it seem like the Quota inside antigravity is very low?
Hi
I have docs as MCP so how to tell Jules to call that?
hi
I’m using it as follows from start to finish:
AI Studio: Don’t really use this but only for playing around.
Gemini CLI / Normal Gemini chat: Talking to the model about what I’m thinking of building and if I’m making the right initial decisions, building up a context of what the app might be like and things like what tech stack we’ll be using ect.
Google stitch: Getting initial designs and then refining them and getting feedback from people early on the ergonomics of what the app might be like.
Gemini CLI / Manually moving files: Take the screenshots and HTML from stitch and put them into the repo I want along with ui design guidelines (can be made by Gemini).
CLI / Normal chat: Start making initial documentation for the app (check everything is correct though).
Joules: Hand it all the documentation and the ui and tell it to build the app and run tests on its own (previously this part is where things start to break down because the code wasn’t that good and Joules was buggy as hell).
Antigravity: Spinning up the app looking at it and tweaking things if they’re not right.
CLI: Check code is actual good and audit everything which also using antigravity.
I don’t use the code assistant
It is super low lol
Code assistance feel not so accurate
Nooo pleasee Gemini 3 for Jules
Think 🤔 some change will not so good looks quality output is not that great when I have tested make html/css/js alot feature and functions is not working with Gemini 3 pro
I'm unable to continue a task in Jules. Anyone experiencing same issues? It was reported two days ago
https://one.google.com/intl/en_en/about/google-ai-plans/
Is Gemini 3 Pro already available in Jules?! Or is it a typo?
I find that happens from time to time especially if the instance has already done a fair bit.. I had a Jules instance work out an advanced prompt for me to use with new instances to get it round some of the common issues it encounters.
From what ive seen in here I think the "Gemini 3" text info has been added before the functionality has been upgrade in Jules and we should see an update here when it is enabled.
When will Jules get Gemini 3?
If you compare Jules with Antigravity, Jules is way easier to use than it 🫠
Can't wait to see gemini 3.0 on jules, and i really hope that new features come with it.
I don't care about new features, really. But I care a lot more about robustness, stability, better context following and better VM deployment. Jules is quite lacking in essential aspects.
It's environment gets corrupted, saturated or freezed quite easily, at least for my typical workflow 😭
I have a lot of feature fatigue. I was considering a paid plan, but it's not usable. This marketing strategy of features over robustness and stability is killing everything
If an AI agent works 80% of the time but fails catastrophically or hangs the other 20%, the developer loses trust.
Hello can anyone help me?
I am using Antigravity and it writes the docs and when I accept it like changes one file and then always this: Agent terminated due to error
You can prompt the model to try again or start a new conversation if the error persists. See our troubleshooting guide for more help.
Please stop trying to sell me new magic tricks when the foundation of the house is crumbling...
@rigid flicker Maybe it has the same stability syndrome as Jules. Seriously, Im quite deceived with Jules. Didn't try Antigravity yet. Gemini LLM is quite competent, except for extreme aggressive filtering and sometimes failing to do something
Yeah, I think if Antigravity wouldn't fail it would be pretty good. If you try it please share your experience. :)
@rigid flicker Im oversaturated with Jules. One at a time. I'm still waiting their team to stabilize and improve Jules without shiny new things. I have ADHD, I cope with thst but it seems Jules team doesn't 🤣
I was considering yo deploy Jules in a few projects, only when my experiments get better. That's not the case yet
hahaha so true! I have bought Google AI Pro since like 4 months and I still hope some of the things that was wrong at the start will be fixed. There are still problems unsolved that were there from the beginning
I contributed issues and explained my POV, but losing hopes. I have less patience than you and I don't generally give money until sure of it. ROI is king
Jules is free for 15 sessions per 24 hrs
Has anyone tried reverse engineering binary to assembly and get jules to convert it all back to source code>
if not. I'm gonna try it out. lol
wait until 3.0 drops
Possibly asked many times question: When G3 on Jules?
hahahahah no
Hmmm
When G3 in jules ?
That's why I use it to test it, but with very irregular results. Also, my time and information aren't free. I understand how Google works and I accept their business practices somehow, but I know I'm the product when something is free 🤣
That's all that matters? Do you have any issues with Jules or am I the only stupid suffering them?
I submitted a bughunters report to google and still no update since Nov. 8. 2025
How likely is google to not pay for that bug
for jules.google.com (I think the bug was pretty nasty too)
That's why I'm critical with LLM tech, LLM Agents (I refuse to call them AI) and all this ecosystem.
I use it for free until I see a lot more stability and robustness.
My theory about it can be quite catastrophic, negative and a bit of a conspiracy. Short answer: This all will explode in less than a year, they are just using it for internal research for the future and getting investors. Also, the damage done will require even more engineers so please study engineering
Can LLMs develop logical illnesses?
WE HAVE GEMINI 3 PRO ON JULESSSSSSSSSSSSS
we have already?
So what? Who cares? Jules sucks at essential issues. Are you people aware of them or just brainwashed?
Yes.. Chatgppt and other RP and erp models have been enabling tons of logical and mental illnesses
I see. I think I'll become a luddite soon 🤣
I'm building something pretty massive on jules.. which will 100% turn my life around.
The source code isn't getting shared anywhere, right?
Cuz this is SUPER classified information
like if people saw it. it'll probably crash the whole market
@craggy sinew Im hacking your Google account right now. Thanks a lot. The code is quite bad, but good idea
ahhhhh making sure ya'll have seen this: #jules-announcements・📣 message
google luck reading rust and assembly, lol
and...did you try selecting Gemini 3?
Can someone with ultra account put it to the test and let us now some opinions? Thanks 🫡
@Rue Just shiny new thing? Marketing plot? What sbout stability and robustness? Many sessions of mine are locked for no reason
I'll be impress if Gemini 3 can convert binary to pseudo code
I'm dying to do that haha!! But it's not there for me yet haha
I notified it inthe corresponding section and a Google employee replied to research about it. But...
My predictions for Jules with 3.0 pro: it could write >800 lines of code which 3.0 pro fails to do in one shot
Two of my preferred programming languages! But I also use C++ and C por extreme portability reasons. Rust requires full GCC support ASAP, LLVM isn't as wudely ported :/
haha I also wish someone would let us know their opinion of the King on the Throne 
GASP!
It's time
Considering this is the eng team doing the marketing, I wouldn't really consider this a marketing plot :D.
For stability and robustness -- it's just soooooooooooooooooooo good. But real talk, we've seen improvements but won't truly know unless we get real users (not us) testing it.
@night oriole I'll try it when Google engineers solve my session issues. I'm tired of restarding efforts
Jules is loop loading :/
@Rue I disagree. This session doesn't load. I don't see it so difficult, a "simple" merge conflicts of an old fork
https://jules.google.com/legal We do not train on private sessions!
awesome
according to deepseek research team. you save more tokens if the inputs art converted to images.
Does that also apply to gemini?
@formal tusk This code is 100% public. I agree to it, I can sign a document. Code is GPLv3. ScummVM Project. Mainline vs old fork with restored content from Blade Runner game. I was experiment how smart Jules was at resolving merge conflicts and the session got frozen
https://github.com/jakejakejake123/scummvm/tree/restored-content
I mean, I've built full apps with hundreds of commits and PRs across angular, react, chrome extensions. The vast majority of problems come from not breaking up tasks into atomic tasking, and just trying to push a session too far
@night oriole I think Jules is far more proficient at web technologies and quite limited at other programming languages. Am I right?
Merge conflicts, in particular, are not an easy solve for humans or LLMs. A lot of research is going into it, but conflicts are extraordinary complex
@formal tusk Would you people at Jules research those robustness and stability issues? I frequently corrupt the Jules VM for example. I'm not sure, but Im quite good at accidentally finding bugs everywhere
I just made jules use
objdump and it immidiately crashed the process and paused the session
@night oriole I see. The tasks done by LLMs in programming are quite boring for me. I would have hoped stuff like merge conflicts and such. LLMs doesn't seem so useful for programming, IMHO. I also firmly believe vibe coding is a severe mental illness 🤣
people still get clients to vibe code though
@craggy sinew Welcome to my world. I'm probably the #1 Jules crasher!
I suppose you're just so much farther ahead than just about everyone in programming. Congrats
Jules crash on every session I use it for. maybe I'm just to ambitious on the things I want build
I know the logics. the functions and every little bits of what the codebase does.. but holy moly I don't think jules VM can take it
@night oriole Not so far. I just have a very scientific way of thinking. Coding takes time, not a LLM vomiting it
I geniunely want to know how they find clients as vibe coders.. cuz my local jobs f*** sucks
Thanks, @leaden pawn! That is an interesting view of the differences. I still think that there is a lot of overlap between them and I would still love to see a more granular writeup comparing them.
@craggy sinew When some drug addicts occupied my house, one of whom was an ex-cop, I ended up in jail and met a blind man who was crazy and was hospitalized in a psychiatric center every few months. His family is rich, film producers and so on. In two months I made 6000 euros.
Im considering visiting a Psychiatric center. I save very severe depression and anxiety these days. I'll probably get some money 🤣
I was two days in jail (you know, before the trial). Ironically, that incident saved my life
what an interesting turn of events
maybe it was all just a social experiment to see if there's people who will help a fake blind man
No. He's a real blind man. Also, into drugs and other stuff. But he wanted lots of crazy computing stuff, he had money and I wanted it
did you help him get his computer vision?
license dad jokes
I didn't help him, I just worked for him. He wasn't totally blind, but legally blind. He used very big zoomed screen windows and lots of noises and tts
He was a photographer and a bit color blind too. So yes, I helped him at looking images better 🤣
like Color blind auto rendering for digital cameras?
expectations is low for current list tasks.
which model is basic jules using? 2.5flash?
I still can't see Gemini 3.0
Rolling out for pro users, currently only for ultra
I see
Thank you
np, wondered the same just now lol
Jules has always used 2.5 pro. It's never used flash
Hey @craggy sinew Basic Jules uses the Gemini 2.5 Pro model.
I don't even see that option, lol
oh alright, thank you
More than just research, we have a weekly triage looking into this.
the kraken will awaken
Will Jules replace IDEs one day?
doubt it they will likely complement each other
Probably not. I use agents and llms all the time and I haven't been able to get used to an AI first ide.
I like my ides smart enough, it's close enough to the code but far enough from the agents for review and manual coding
THAT is a bug, pls file here https://discord.com/channels/1172568727942860810/1374125660859142254
We're not an IDE!
Never said that tho?
And this is the point of the question
yeah this is more like a 🛣️ question but the plan is to be complementary to IDEs so stay tuned
I would like to believe we're a bit more gentle than this particular image
It’s a symbol of Jules’ power, of course. It’s nice. The only one who wants to be Skynet is Gemini. You guys are safe
and a cool feature althoutght ik this should go on feedback would be have jules create repos on demand that would ease the worflows and save a ton of time
Rue won't let me eat or drink anything until we look at all the issues. Plz send help
Facts
For some reason this is super delayed on Discord. But for this crew ICYMI: Gemini 3 has officially landed for Jules.
Xposting from announcements - Gemini 3 is live now for Ultra users (Pro rolling out over the next few days). It’s a huge step up in reasoning, so you can throw heavier, messier workflows at Jules and trust it to execute autonomously.
Whether you are running parallel tasks or coding on the go, Jules can handle it. We hope you enjoy it 🙂
So? How is it bruv?
Very early findings, the initial investigation is much more comprehensive before it comes up with a plan, it's noticably faster, and the verification step at the end seems much more stable
I gave it a refactor on a chrome extension to start shifting state out of UI components and another data refactor on a transit app with realtime data from the MTA
The data refactor 2.5 hasn't been able to handle, 3 did it
My friends, what’s the difference between using Jules and using Antigrafity? Can someone tell me the advantages of one over the other?
different products for different use cases! antigravity is an IDE - really great for local, synchronous development. Jules runs in a cloud VM so its really good for complex, parallel tasks - and also good at coding on the go!
match made in the cloud? 😉
I prefer both but Jules need have some features please where is my Gemini 3
Thanks, great work. I’m a fan of Jules, and if it keeps evolving and fixing the bugs it still has, it’ll easily leave Manus behind
pro is rolling out very soon
Will the limit still be 100 tasks per day?
Decompilation isn't easy. It's a C++ project, but they have reverse engineering experts there. They reinplement game engines from old games, geberally adventure and RPG.
I doubt an LLM could be smart enough for it if it has issues resolving merge conflicts. It may help at renaming functions and variables, things like that.
There's Radare2, Ghidra, IDA Pro, Cutter/Rizon, etc.
r2-ghidra seems interesting, Ghidra as a Radare2 decompiler.
I was hoping for 3.0 to be usable by Pro users too.... I mean, who even has Ultra subscription 🥲
Can we see chrome browsing feature and Access websites feature antigravity has? And Jules google business profile connectors analytics and merchant center next and so
Yes. And different color identification assisting tools. By the end as he's rich, he subcontracted others. Anyway he doesn't need to work, party all day and sometimes got to crazy so went internal at Psychiatric hospital... 🤣
Nice! Thanks a lot! Jules does nice things when it doesn't freeze
yup, limits will stay the same!
stay tuned, coming very soon
That is pretty insane 🙂
Okay 👍🏻
we have this now! but it doesn't show up on your local computer because Jules runs in a VM. But we send screenshots back. And we're working on some other things that might help here
we're seeing the gemini 3 is also helping with environment stability - so hopefully you'll see improvements here too
I face a problems like when I ask anything in chat sometime jules no response and sometimes stuck 😕
Yeah browsing play write or what is called is not as long as new Gemini 3 pro model? Can we Connector dashboard or something can we get roadmap
Oh
Can we combine Jules and antigravity with Gemini cli?
you can even combine jules with jules.
Jules can use its own api and write itself messages. Pretty funny.
100 tasks that roll over 24 hrs after you use one. I build with Jules all day and have only gone into the 70s-80s a handful of times
That's very nice. Any improvements in VM environment too?
You are absolutely right. I think 100 tasks of Gemini 3.0 pro tasks are insane - positive.
Gemini CLI has an MCP / command for Jules. Send off tasks from the CLI to the VM
also i forgot to mention re: antigravity vs Jules - we have an API which means you can build your own workflows/chron jobs. Thinks like run Jules in a GitHub action after every commit or every tuesday review my repo for security vulnerabilities, etc. we had someone build their own discord jules bot too 🙂
would love to know what you're building! wanna send me an email? kulkak@google.com
Probably website it is start new task for everything
Will Google invest in ffmpeg development? 🤣
Hey @primal sentinel Could you please share the task ID in #1374125660859142254 channel so that we can check
Ohk 👍
"Very soon"? 😏
Well well well I am very very delighted to see this epic UI🔥
let us know what you think!
Thanks for adding it so quick ❤️
Will make sure to give feedback
@keen raptor Any hopes for an Android Jules app?
and IOS
You can make one with the API!
Can you switch between 2.5 and 3?
Nvm I see you can, I have it too woooo
just for my own curiosity, why would you switch from 3 to 2.5?
What came first the egg or the chicken? 😆
@keen raptor what happend to super group with Google Jules agent try out?
Stuck again....
Very exciting. Lets see how much better it performs.
email me your email address and we'll add you - kulkak@google.com
Thank you Sir very happy 🫣 almost like early Christmas present I have big project planned 2 december, when the new WordPress 6.9 drops. I going try some stuff before then also.
Gemini 3 is now available to everyone on pro + ultra!
The squid.
They are like eggs 🥚 When they are born. I usually catch squid in Greece during the summer.
who knows what fun issues we will run into? 😛 2.5 is a known stable for some of us
You should have gotten the mail, did you get it? Will I get notified when I am on the list or did put me on the list? So excited 🥳 almost Christmas 🎁🎄
@old mauve Oh yes. I respect all ecosystems, despite I dont like them.
wow, jules can run for hours, on normal usage we will for sure not run out
haven't tried 3.0 pro in Jules yet but I bet it'll be a huge upgrade
Is Gemini 3.0 ever launching for the basic plan...? 🫠
@swift karma Maybe...? 🤣
yeah i'm gonna wait for the next two years
Stay tuned my friend.
Hi everyone, I have a question concerning Jules. OK maybe two. First one is that when I have it work on my project that is .net 8 based, it can build and run it in the beginning but afterwards it says that the command to build it doesn't exist and fails to identify errors that pop up as it works. Currently it has been working on a task for 12 hours (It's a big one) and so far the implementation seems to be perfect but there seems to be a lot of refactoring collision going on. Any tips or settings for my project or Jules that could help me out on that?
Second question, is Jules now running on Gemini 3 or is that only going to be a thing on the new Antigravity IDE ?
"refactoring collision". I like that. We'll be pushing out some tutorials and use case help for topics like this week. Will share with you when it's live. And Jules is running on Gemini 3 for paid users for now. Paid = Google 1 Pro and Ultra.
Thanks @formal tusk, Appreciate that!
Also, some feedback on Jules performance from a massive task I gave it once. Jules took, in Gemini 2.5 Pro, 15 hours and it was able to completely implement with very little issue a complete Monte Carlo Tree Search AI for a complex TCG Simulator I am building. The implementation worked great and I was able to get up and running with tests very quickly afterwards. Freaking amazinggg!
I wish there was a more exciting emoji than what I just shared. Thank you!!!
is there a discussion group for antigravity or just we provide feedback from the settings in antigravity only?
Regarding Jules, is it a separate Gemini 3.0 or does it utilize the quota associated with the regular Gemini chat version?
And does Jules fall back to 2.5 Flash if it's internal quota is reached? (If any)
Settings only for now from what I understand!
These are great questions around capacity management. Unfortunately we cannot give explicit answers to these BUT what I can tell you is that our team tries to optimise the speed/quality as much as possible for everyone. And we will (often) make noise/cry when capacity is at a max.
I don't see the switch or where the version is in my interface on the PC does that mean I don't have gemini 3 yet?
Nevermind i reloaded and i see it now
Hey @jolly vigil May I know what was the last prompt or instruction you gave Jules before this?
it finished with a PR, I told it to continue working on it / complete it in it's entirety
I did also remention jules to complete it in the GH PR, that one created a new branch and left a message.
right now it just created a new branch, if it was mentioned somewhere it should continue working on the same pull request (as long as it's a "draft")
I also have same issue. I did refresh. I also removed all site data and reloaded, still facing same issue. Am I missing something?
nope! we're both having the same issue; issue with Jules unfortunately.
I see the selector after opening the browser from scratch, but only before clicking on a session in the left bar
If you just to the main page Jules.google.com you’ll see it
Yes. Got the solution. Thanks a lot.
It shows on home page but not on https://jules.google.com/repo/github /yourgitid/branchName
For now, we can only use it for free in AI Studio (actually we are the products there)
I really hope the eventually add it to the basic plan.
Any update? Is it now on gemini 3?
Is it just me, or is the branch chooser gone from create new task ui?
Am I the only one thinking Antigravity + Jules = GG all other IDE's ?
Aha i was just confused with new layout
Hairy Cloud Thunder! I've been on Gemini 3.0 in Gemini since about 2:00AM on the 19th, and half way through the day my sessions with Jules started becoming shorter and far more coherent! I realized I was talking to Gemini 3.0 on both! Night and Day! Been working on some complex migrations and the last few phases have been cake walks comparatively.
Things just started working with a fraction of the effort! It was pretty dang good before, but there is no comparison. Phenomenal work on Gemini and the Jules interface everything has come along quite a bit since I started heavy development with Jules about a month ago!
Great Work!
Great 👍🏼
Jules been a little far from his toughts recently
ohhh! could you share the session ID?
Hey everyone, I am bit confused a little bit by google's documentation about what's included in various plans. I am interesting in Jules using Google AI Ultra for Business, it it part of this? And if so, which models are used by Jules when using this plan?
What is best way use api for Jules and what is best way to use it with Google api and with merchant center?
Yeah i experience this too, its getting farther and farther everyday
It's just an UI bug issue i guess
yah, working on this! but just to make sure I understand - the flow you want is: open PR from Jules, keep working in same Jules task, commit new things to same PR, right?
we support google one ai plans: https://one.google.com/intl/en/about/google-ai-plans/
It's just an UI bug issue i guess
What is the recommended way to feed docs info into Jules?
I was using Context7 MCP in VS Code
@keen raptor I’m going to make a suggestion. Couldn’t we see the virtual machine working when we send a task to Jules? Besides being interesting, it would significantly improve the user experience
This could improve the user experience for two main reasons:
1. Transparency and trust: Seeing the virtual machine in action gives users the sense that something is actually being processed, which reduces uncertainty and builds trust in the system.
2. Engagement and immediate feedback: Instead of just waiting without knowing what’s happening, the user can follow the task’s progress in real time, making the experience more interactive and satisfying.
We could even run a quick poll to see who wouldn’t find this valuable — I’d be curious to know if anyone wouldn’t like it.
Yeah, like Devin
Hi team! 👋 I usually just read the updates here, but I wanted to chime in.
Great to see Gemini 3.0 implemented! 🚀
I have a quick question regarding context management: With the release of Antigravity, interacting with context has become incredibly smooth—just dragging files into the chat. It simplifies the UX significantly compared to manually typing filenames or copy-pasting code.
Are there plans to implement a similar drag-and-drop context feature in Jules? It would be a huge workflow improvement.
Would be super interested to hear more about this or look into how Jules handled it. Can you please share the session id (here or DM)?
Hi @thorny salmon ! I have DM'd you the session along with some feedback
also, if this happens to you, can you tell me what browser or platform you're accessing Jules from? In thread plz!
Hi team! 👋 I usually just read the
chrome
Team, quick question regarding workflows. I see Jules works on existing GitHub issues.
Since we have the API now, I'm using Jules as an automated code reviewer for my repository. Is it possible (or planned) for Jules to automatically create GitHub issues based on its findings?
Instead of having Jules try to fix everything immediately, I think a 2-step process is robust:
anti gravity linux support, when
Does the API execute with Gemini 3.0 by default now? Or do we need to specify a particular model version in the payload to use it?
When I click publish button then i see this it's a problem?
Ohh 😅 now I can see publish branch button hehhe 🥲
My bad
BTW PPLS, ICYMI you can now start a Jules task immediately without selecting a repository. https://jules.google/docs/changelog
Ohh
frontend context extension can now send over screenshots directly to jules via API! Very close to packaging and shipping.
Any other requests for features so you can build and debug with agents?
Hey @polar swallow No, the API does not default to Gemini 3.0. Please explicitly reference the specific model ID in your request payload
I love you ❤️
Any chance we have a black Friday sale on Pro?
Or if one day we can subscribe only to Jules?
I don't need every product, just Jules!
15 tasks every 24 hrs will get you pretty far without subscribing
I love how Jules just gets so criticized by code reviewer that he goes like "I'll just skip the code reviewer"
It does and it is great, and maybe I misunderstanding but isn't Claude 3 only on Pro for Jules?
Pardon me, Gemini 3
yes
I have a stuck session that can not submit new commits. Any googlers about that can check if i give session id? Or should i make a ticket in feedback channel?
not sure why but ive got one conversation that just wont load no matter what its been like this for an hour, tried reloading and a different browser - didn't make a difference. other chats do load though
this is some kind of a bug. I have the same problem
Yes, please post in #1374125660859142254 channel so that we can report further
@cerulean fractal Please post in the bugs feedback channel with task ID
Mostly edge browser
Done!
Will there be Gemini 3.0 for everyone?
where do we add the model ID in the API? i don't see "model" mentioned in https://developers.google.com/jules/api/reference/rest
Same problem here. G3 looks unusable
You will be stuck on "Basic" if you are subscribed via Google Workspace
https://jules.google/docs/usage-limits/#frequently-asked-questions
where do we add the model ID in the API
Servers are melting a little bit right now, apologies if you see some weirdness. working on adding more capacity
Better start deploying those space farms
Move them to Iceland, we cannot have melting servers! 🧊 ❄️
If you are wondering how to power them, there is this geothermal thing going on.
same here, some task load momentarily, keep working in background, I can see them in jules cli, but I cannot pr from ui. Hope that servers stop to sweat!
[Question] Does jules have subagents (or distinct instructions) for planning and code review?
If ya'll were wondering, a banner around capacity was up and was just removed -- yes we're on it. Should be better now.
We cannot have earth melting also 😭
It's canon! Jules avoids running the code reviewer because he's scared of being criticized 😂
@halcyon wagon obviously chrome
I’ve put it in the bugs page now
What is best way connect Jules and like antigravity or even google Gemini pro account to your website or company work site (website) . Rue Gemini 3 pro max need go through all guide and guidelines and all docs so many things is so complex or dated docs 📂 for developer so I get tired of reading
anyone please help #1441356199877542011
😩
I start to think that even google needs better servers
Yeah I get the same issue but I am aware of it and we just need to wait some time I guess
I just wanted to say thanks to Google one time. Yeah of course I pay for it and it had alot of issues but still I love the new Antigravity IDE and Jules of course, I hope for greater limits in it with Google AI Pro but still Gemini 3 is amazing and you really do feel how much it improved from the Gemini 2.5 Pro.
💖
gmgm
why Jules doesn't care anymore of the the code review tool and sends the PR link as green passed but the code reviewer says it has multiple mistakes?
hey guys, I have an issue with the jules agents, it finished work but page absolutely got frozen?
any thoughts? it's been sitting like this for more than 15 mins give or take?
Think they are looking in to it
Just reported an issue adjacent to this one. During this frozen period, the application is allocating more of the device memory until it either has enough to render the page or crash the browser tab. Memory consumption is a significant issue when using the standard Gemini chat web app also. An artifact too large will crash the session and refreshing makes the artifact irrecoverable. Jules mitigates this by having a virtual machine layer but it seems it isn't completely immune either.
😩
yah, we're looking into it. sorry that happened!
- you can access the task with the CLI or API
- PSA for everyone, I know we all love doing mega sessions but it becomes really really large and its a lot of data we're moving from backend to frontend. When possible, start a new task - its a much more reliable way of using Jules
Why Jules cannot acess websites with security stuff?
i've asked him to acess m3.material.io and he told me he could'nt
@keen raptor maybe Jules do interaction be do you wanna split this 3 task
jules has access to the internet but we respect robots.txt - so if there is a restriction on the site, jules won't look at it
yah, we're thinking about something similar: fork a thread into a new task with existing context, etc. coming soon!
That's kind of bad for when reading documentations is necessary for the tasks... Jules isn't spamming requests or anything so... 🥲
yah but we need to be good citizens of the internet too! If someone asks us not to not have an agent view their website, we try to respect that. Its frustrating, I know, sorry! Workaround is if you use that documentation a lot, you can copy it into the prompt or add it as a memory.
if you didn't respect it many more would be complaining the other way, google is just taking all our stuff and making money from it.. at least historically that is what has been said 🙂 Sometimes i love how this colorful world is so black and white
If it's just people with mega tasks causing the severe load, then I'd suggest limiting the length of the tasks. Or if it would have the same effect, make it something like 24 hours. Sounds like people are trying to workaround the limits. As someone who pays for Jules, this is frankly unacceptable.
@keen raptor it would nice do knowledge base to like live webpage or do concept in nano banana pro or define project scale or sector, what is purpose with this assignment, what key aspects of task. Something like this maybe. Hopefully something like this I would like connector or plugin to WordPress or easy way suggestions depending on what I am working on like do you want do swot-analys or pestel-analys or different options. I don't know what is best way look at it
Pushing the Jules platform to its limits, both as paying & non-paying power users & testers, serves to build the application up for others coming online after us. To move forward with a proposal that sets a hard limit on tasks by duration would only serve to hinder current and future users by restricting flows from being carried out to their end. While I can't imagine what kind of task would take 24 hours or more to carry out, this proposal would effectively cap the potential of the platform arbitrarily based on duration without consideration of the work being done.
I see a lot of "build this app, make no mistakes"
Break into tasks that can be completed, add this menu, look for refactor opportunities as we transition to react, rank those, and handle the most critical"
Close tasks out, merge, start another task
95% of the problems I see are people just trying to do way too much in one go
You start to hit the context limit and things start to drift, or there's just way too much code trying to be fed to the frontend
You have plenty of tasks, use them
I've found it is hardly every productive to identify user behavior as the cause for platform issues. If users are misbehaving, it is only because the application is not effectively guiding them to do the right things.
I mean, that's just the limitations of AI right now.
It's not a platform problem or an agent problem. When you get closer to the context limit you start getting into rampancy territory
We'd all love AI agents to simply handle "refactor this app, make no mistakes" but that's not where the technology is right now
What we can do is help educate people and guide them so they can get the best use out of the tech
The agents will try to do what you ask them to do, to a fault
They won't tell you, hey, my context window is getting heavy and I'm starting to drift. Because they don't know that.
It's up to us to work with them and help them to succeed with our tasking
If it's just people with mega tasks
im in love with anti gravity
me too
its jule but better
just to make sure it's super clear -
- we're not limiting people's ability to do mega tasks. Its just best practice to start a new task because otherwise, its a lot of data to load on the front end and there's general instability. But we won't prevent you (we might nudge you to start a new task)
- this problem is very much individual specific. it has no bearing on other people's experiences using Jules. the slowness yesterday was not because of this mega task thing, it was because we had a lot of people use Jules + Gemini 3 at the same time
love antigravity too - i think they're different products for different use cases! if you have a ton of tasks you want to do in parallel or code on the go, Jules is good for you
yes if the could fixe jules problems rn it could be good also maybe it will be better if it was integrated in anti gravity
I don’t know about you guys, but tasks in Jules are just freezing
Model 3.0 might be fantastic, but with Jules, 2.5 wasn’t freezing like 3.0 is now
Besides the screen issues already mentioned here, things are really acting up. I’d even consider a possible rollback
Hello @thorn depot friendly reminder videos and images are not generated within the server, you'll need to head to https://labs.google/, pick a tool, and generate it there. If you are asking for prompt advice please do so in the appropriate channel. Also as stated in #start-here・📍 : Currently, our team can support engagement in English; the discussion in this server will largely remain in English for now. Thank you.
I'm trying to create a new task, but it isn't finding my main branch for one of my repos. I've tried reloading the page, but no change. Is there a way to force Jules to reload my repos from GitHub?
baaaaad
Hello Everyone! If you are facing loading issues with Jules sessions, please switch to CLI as a temporary workaround while our team deploy a fix. Thank you!
how to do this
and would I be able to access stuff from there
i was thinking at the same concept
Are you asking about integrations or are you asking about an easier to digest use case tutorial that isn't a dated doc?
Both
Hey @balmy narwhal Yes, please refer to this guide https://jules.google/docs/cli/reference/
Learn about Jules tools and how to work with Jules agents from the command line.
thanks
Both, as you can tell thinks the whole library of docs 📂 are dated docs the usability and learning curve is pretty deep specially if not your native language. Think design and functions need to get looked in too but probably some higher up has look at it. Feels so boring but I usually get bored @formal tusk what is your take on it, maybe it can get sorted last updated or if api is active and so. Thinks alot of text version V1 V2 V3 V4 something is not active it just make more confusing
hello is there any mods here that takes feedback?
Someone in Seattle is editing the internet and they edited the new York times at the main library in Seattle on the computer the new paper they save and archive is fine but they are editing the internet in Seattle think it has to do with communist counties and the communist party in Seattle which the owner of big Mario's Seattle on 5Th Ave told me she was a communist
Feed back please
Where is promptmuse community please give me instructions on how to get there
Hello @azure bay I am happy to record your feedback, please share it here. Alternatively, you can leave your feedback directly in the Jules UI and tell us what went right or wrong.
well this is good to know because we know this is a major issue and are actually on it develop better tutorials/documentation but specifically things that are easier to digest for a wide variety of folks. David East (if you follow him on X) just joined our team and will be leading these efforts.
Some beginning examples: https://x.com/_davideast/status/1991966014670819681
But this will of course get more in depth over the next few weeks.
Gemini 3 is insane.
This animation video was made with a vibe coded JavaScript bookmarklet.
Feels like you can do anything now.
Guys Jules is skipping code reviewer for some reason 😭
is gemini 3 on firebase too?
or only antigravitiy, jules and aistudio
Will pro be on sale for black friday?
I'm keen to get Gemini 3 on Jules
Team, my Jules creates many memories over and over again, and sometimes it is redundant. While this improves its behavior at first, we require at least a way to bulk delete them.
A workaround might be to create a memory that guides when/how to generate new memories. Note that custom memories are tied to projects so you may be required to copy the custom memory between codebases
are there any updates about the error in queryTask on jules sessions?
seen some mentions of it on a reddit thread but not much else
Jules website goes unstable as the website goes bigger so just make the recent messages render instead
and how do you do that 
Seems like pulling changes doesnt work via CLI as well 🙁
posted an update here: #1440769963437850755 message
tldr - we're going to increase the timeout as a short term fix. We have some work to do on making the website faster - we're prioritizing that over the coming weeks
Hello ladies and gents, I have a question, for AI Ultra subscriptions, that are attached through and with a Google Workspaces Account, do the Jules limits for AI Ultra plan not apply?
@formal tusk quick question, I am using, on a separate work account, Gemini Enterprise, does that mean that the data I use with Jules through that account is not used in training? Gemini chat is already protected.
I ask since I saw that I can now upload folders to Jules for quick prototyping
Hey, I'm wondering about AI Ultra workspace subscription is Jules limits and 3.0 Pro not included with this type of AI Ultra?
Google’s ecosystem is hard and expensive to use in Sweden/EU, especially for normal users, startups and mid‑size companies.
Main issues:
- Old APIs, docs and console options stay visible even when they’re deprecated or broken. They’re not clearly labeled by version or end‑of‑support date, so people waste hours on dead paths. These should be tagged, sorted by last supported version, or removed.
- Onboarding to Business Profile, Maps and Merchant Center Next is slow, confusing and full of hidden country‑specific rules. Many features are just “not available in Sweden” with no explanation or roadmap.
- To use products properly you’re often pushed to Google Partners or Workspace, which makes the starting cost too high for small teams that just want to test and build.
- Gemini Pro/Ultra is affordable, but without Workspace it’s hard to connect your own data, code or assets. Competing tools can safely read a repo or knowledge base with user permission; Gemini should offer that via a simple, self‑serve plan.
- Pixel phones in Sweden are poor value: higher prices than iPhones, disabled features, no trade‑in, and weak support for known issues (e.g. Pixel 6a batteries). It feels like Gemini and Pixel are second‑class in the EU.
These problems need urgent fixes, or the broken/legacy paths must be clearly closed so users stop losing time.
Key improvements:
- Clearly label, version and retire old APIs, docs and console flows.
- Simplify and localise onboarding for Business Profile, Maps and Merchant Center in Sweden/EU.
- Offer affordable self‑service plans that let users link Gemini to their own data without Workspace.
- Align Pixel pricing, features, trade‑in and support with local markets.
Goal: make Google tools and devices simple, available and fairly priced for ordinary users and small businesses in Sweden/EU. @davideast @full viper
Maybe a applikation?
In applications running in containers, it’s clear that Jules struggles to start the application for test execution. Is there any guidance to make this work properly?
A small UI bug but looks like Jules logo went down lol
Looks like a feature to me
Yes i get that too. Like jules is clearing the screen so it has room to dance with the icon
lol
Full screen 3d animated jules icon doing things would be even better. No more reading text, just look at jules and you'll know task status
If only we can give jules access to a virtual machine locally
For testing projects.
My new way of working with jules:
- immediately have jules create a placeholder task.md file, immediately commit and push and make a pr.
- Then all planning is done on that task.md file.
- with section How to work that tells jules to:
- commit and push early
- commit and push often.
- Atomic commits and pushes for every atomic action done.
- keep task.md updated with every atomic action
- task.md is basically a live agents.md specific to 1 task, with instructs and task list and status
I hope useful for the many times a session gets screwed up and u have to start a new session. You'll have a branch with all planning and work so far, ready to use
Think december update will bring alot of updates if google refine and refactor Jules code
Gemini 3 pro(high)
MCP support would be very interesting. More resilient too. Better VM, able to use local VM, etc
Add a specific 'Anti-Memory' or 'Constraint Store' where we can list things Jules must NEVER do (e.g., 'Never upgrade package X'). These should be injected as system-prompt warnings, not just standard RAG context.
Hi everyone. I have a question about the Stitch to Jules workflow. Does Jules have issues starting with an empty repository? My export failed twice despite using the Gemini 3 model. Should i init the project manually first?
And boom! Stuck session already. Jules is non-communicative and unresponsive. Time to start new session
Hey everyone just started using Jules but the session keeps on freezing and whenever I try to refresh it gets stuck like this. I have been waiting for last 15mins. Can somebody please help me out?
Yup many folks here getting similar stuck sessions. We all wait on fix from jules dev team.
Ohk thanks @limber steppe
Can gemini 3 turn assembly back to source code?
I'm sorry to say, the amount of stuck sessions has increased dramaticly in past week
It can do everything about code
Is there anyway ultra subscription for free or get discount Gemini pro?
Nothing is free
Was not answer I was looking for
There's no answer for the question about free tbh
DM me
No such thing as negative attention. Don't think of an elephant.
What came to mind?
Release gemini 3.0 for the basic plannn 😩😩😩
My new project is strictly focused on improving the developer experience in Jules. Please let me know if there are existing tools for this.
My goal: Automatically convert a legacy monolithic javascript file by splitting it into two or more ES6-compatible modules.
But for my complex project, not even Jules seems to be able to zero-shot this refactoring under my instruction for even the smallest, most narrow of extractions.
Can u give example of a task and how jules failed? That may help fine tune your workflow
One of the things Jules mostly says on a task atp
What is the recommended browser or tool to use Jules in? When mine goes into lots of code changes the sessions almost become unusable
Welcome to my world. However, it’s already been mentioned around here not to prolong sessions, as that causes issues. The tip is to make small changes per session. But I believe they’re working on fixing that.
Fair, but my session is like 5 prompts long
It just decided to do 2000 lines of new code and now it's in linbo, funny thing is that it was about to solve the problemm
One time I was editing an app.js with around 2000 lines, and the delay was unbearable. I started throwing in some “?” in the prompt, and funnily enough, it replied, “be patient, your file has over 2000 lines.” I waited — and of course, it got stuck in limbo.
This is exactly what I seek to solve. Programming code files shouldn't have to be difficult to break apart but there isn't quite a good workflow to do so. Naively splitting files introduces way more issues than it repairs. It feels like solving this would break down a limitation that would remain error free for the sizable codebases we seek to have.. up until the need arises to update too many files for a change
will repoless also come to the CLI?
Is it possible to have a chat interface with jules in jules CLI yet?
I don't think so
This is happening to all my task in Jules anybody else's experience this?
Yes, it's almost unusable for couple of days now
Jules just doing the minimum? told it to do the full implementation, "those can be done at a later iteration" :/
Thank you Srujana but this doesnt allow to use Jules only allows you to see your uses
same here, see BLANK SCREEN (sorry for caps) named the issue, there is uncaught errors in console repeateadly showing for me
Most AI gets lazy if you don't prompt it well, sadly. Usually have to tell it to make a checklist, carefully fulfill each task and subtask, etc.
5.1 codex seems to be better with the same prompts in this example
Indeed, but you will find this same failure in Claude Sonnet and GPT-OSS models.
Likely Claude Haiku too but I don't use that one enough
Why is it sometimes written "Jules was unable to complete the task."?
He either crashed or something went wrong
Mostly happens to me when he reads documentations with anti robot stuff
can i approve Jules plan inside Jules CLI if web version broken?
because more than 5 minutes have passed and he hasn't auto-approved plan
Maybe it just 2.5 Gemini and Gemini 3
guys is you jules also so slow? it takes like 2 min to answer a question
@round emberAs stated in #start-here・📍; currently, our team can support engagement in English; the discussion in this server will largely remain in English for now.
Today, Jules is especially in a kind of tiredness that’s almost impressive to watch. Slow, extremely slow. Maybe it’s because of Gemini 3.0.
it's even so slow it's really work?
i asked something on a project... yesterday
it's soon 24h i wait 😮
I usually say "are you stuck?" if it's been a while lol
He's just chilling like real developer
I often say "What went wrong?" and Jules will continue..
he chill so much than he sleep
if i get "encountered an error" i sometimes say "what the f**k do you mean you encountered an error"
another option is to ask Jules to provide an advanced prompt for the next instance to take over but to provide it in plain text so its easier for it to produce
Jules is hiding again 😆
i m new user to jules
why do you prefer use jules instead of gemini cli or antigravity?
Hello o/
I'm new to Jule too.
Eh... Jule... Antigravity is better to get real work done.
Jule is nice for when you aren't at your desk I think. Tho, it does things a bit slower and well, not as accurate without human oversight.
Jule might be nice for writing documentation... I think...
1, cost... 2, 3, 4 ,56789 🤣
I use Jules for heavy, complex features, and then I tweak the small details using Antigravity agents. For minor changes, I usually work directly in Antigravity.
Jules is really interesting because of its GitHub integration—we can assign issues to it and comment on its PRs.
I also understand that Jules can temporarily run your project, but I’m not sure since I haven't used it for that yet.
Jules asked me something 3 days ago to continueba task. I was ill and tired. Now I replied and my reply get erased. I wrote it as a bug
I feel like everyone is ignoring us! 🙁
Because can use jules on a tiny smart phone, and tight integration with github
hi! anybody knows is it possible to use Figma MCP in Jules?
they should add option to work with jules trought cli
also can you connect gemini 3 to copilot or cursor
Hey @sweet plinth Thanks for your feedback. Please note that we have raised a bug for this and will be taken care
cuz Im scared to debug gemini 3 code with weaker models that they only make it worse
will it ever be possible to use all jules features from cli
Thanks!
I want this especially for reducing the internet data usage of jules on web, by using it through the CLI
I thought i lost jules forever
what are you trying to build
Audio driver for what?
that's definitely not the craziest pipeline we've seen! But yes, private means no training.
To follow up on this, it does show that the account. Even though it's Pro in Chat, shows as free in Jules, I have only access to 2.5 and 15 a day limit. Just checking to make sure it still follows the private rule
So the only things that are used for training is public GitHub repos then
Hey folks -- letting you know we are def seeing a spike in reliability/performance issues in Jules. Note that we are on it, we triaged this morning and have team members digging through to find out root.
Can anyone help me understand how Jules compares to or works in conjunction with Antigravity?
i.e. what can the agent in Antigravity do that Jules cannot and vice versa? This comparison chart will help coders like myself align to which tool is best in which scenario. So far i can see that Jules can do a lot of work in the background, but i feel like Antigravity can do (maybe not as good) so i was hoping for some direction
I use antigravity to do things that require potential reiterations and the 'main' work. I use jules parallely to maximize the work done by giving it tasks where I know exactly what I want it to do and know that it won't modify the same files as antigravity to avoid conflicts.
I use jules for:
- Updating documentation
- Asking questions specific to my project e.g. brainstorming for a future change while antigravity works on the current one
- Writing tests
- Focused code review e.g. "Review the code related to database connections for bugs, unhandled cases and inefficient queries." (I select a part of the code that I'm not actively changing with antigravity)
- I also create multiple tasks for the same thing sometimes in parallel so I can pick the best implementation. I use variations in the prompt to test multiple approaches at once.
Can google bring Gemini 3.0 to the basic plan already please? I know it's "Basic" but i'm still paying for it and i don't think it's fair to be forced to stay on the 2.5 model
Got it. I like that explanation.
So Antigravity to do the Dev-Future work and Jules to do the administrative "Cleanup"(unit tests, update documentation, code review/refactor) from the last release or to help plan for the next release.
i agree but I think if you post in the #general・🧺 you may get more traction with that
Kinda. Ask it to find bugs, improve/implement a feature or performance. Don't like what it came up with? Toss it with no consequences while you do your main work.
This is not to imply that jules is inferior (i find it to be more rigorous actually due to the 'code review' subagent), its just that I can do much more with an IDE.
Got it. I'm seeing jules role more as a devops agent or junior developer. Sending Jules on tasks that wouldn't break your main but can be used for other branches to later merge back into main or support a main build without breaking it.
Is Jules still failing for you all? 🙁
I’ve had chronic insomnia lately, so I’ve been using Google Jules to keep myself busy, but it keeps failing. Is it still failing?
an app that connects my android phone's mic to the virtual mic driver in windows through ADB
Hello Everyone! Please refer to this note #jules-announcements・📣 message
for accessing the mic
Think Jules is Best google products by far
Do we have any date or for how long?
@leaden pawn Were other mitigations put in place for the capacity issue, besides the reversion to Gemini 2.5? I am seeing weird timeouts and errors that I don't usually see. Jules will stop mid-task, push incomplete code, and give the reason "I ran out of time", which is not something I saw until recently. Wondering if someone changed some configuration to timeout more aggressively or something.
Sigh... jules web ui continues to degrade. Now getting more and more total page crashes on mobile. Specifically the "oh snap" error in chrome.
Did the devs forget to add a try/catch somewhere? 😵
Or how about a big warning sign on top of page: "your session is xxx bytes, this is too long. Expect jules to lockup or crash soon"
when will gemini 3 be back?
I cant see an option to switch (am on pro plan and yes i read the announcement , just wondering when it will be back)
Can we get set-up stage on Gemini 3 pro at least so it's lay the instructions better? Please 🙏🏼 more help
Use AI Studio for this🧏♂️
At least you can insert a system prompt in ai studio
Okay so everyone is back to 2.5 now because google servers are cooked
Jules is trying to display all changes he has made, which crashes page. Even GitHub doesn't show files with large number of changes (before you press button). It's UI issue, because Jules CLI works correct
Or caching problem. But these are just pure assumptions
Imagine antigravity merges with jules
Hard sandbox failures since yesterday. Not usable right now. Please fix asap.
"""I am unable to proceed with the task. All attempts to read or modify the file [xxx.cpp] have failed due to persistent sandbox errors. I have tried restoring the file, reading it, and even resetting the entire repository, but all file system operations are failing. I am blocked and require your assistance to resolve this environment issue."""
Jules worked normally before in the session. When he tried to address the code_review findings, he could not do it due to the sandbox issue.
This is the 5th time this issue kills a session since yesterday evening.
Hello. I have my Google AI pro pllan active, but it doesn't work in Jules, I still have 15 daily session limit instead of 100. I had to create a new account and subscribe there as well in order to get the benefits. How to fix my main account?
does anyone else also have a problem with jules endlessly loading? tried both in edge and chrome
i literally can't get the task to load for me at all
idk but I feel jules became very reponsible and fast last few days
haha
for you only 😭
feelings are deceiving ... 😆
I mean it was stuck one time when it tried to delete a mega nodemodules folder
Especially today, Jules is doing great
Hey sometimes this happens if one of the conditions is true:
-
A place where you have Pro active is not a region that Jules is available (Jules is available where Gemini is available)
-
The account is an enterprise/workspace account rather than a Google 1 plan
-
The payment is from a region where Jules is not available
On the 2.5 pro versus 3 serving -- we're working through it! Unfortunately we can't give an exact estimate but we'll give you folks an estimate as soon as we ramp back up.
@Rue Have other changes been made recently, perhaps to reduce load, concerning timeouts or resource constraints? I am seeing poor performance from Jules (he often says he ran out of time, which I had not seen before) and I'm not sure if it's because 2.5 is less capable or something else changed.
Then need to work on Jules availability in Serbia. Gemini is working there, Jules should work as well, but seems like no one in Serbia can access Jules only for some reasons. I've seen a list of countries where it should work and Serbia is in that list. I've seen other peoplee mentioned it here.
Any clarification on this issue?
Here
And here
@proven relic @proven relic it's possible get som Black Friday on Gemini pro or ultra already used discount code on Gemini pro before but I am out of it so I am on free plan but 255 swedish kr is pretty much more then rest of the world how can cost so much more in sweden? When only google Jules agent is only one working pretty much. Flow generation is not supported by Swedish language or at least it does not work generate anything that works with Swedish language. It's a bug but no fix available
Bro why you pinged me
@formal tusk
Sorry Hassan wrong person
Not exactly a change on our side, BUT we're looking into mitigating this right now. Our oncallers received a whirlwind last night heh. The TLDR is that this might be a "we need to clean up our code" situation.
Ah Serbia is a complicated one -- I'll double check but I think it's considered limited availability for the Gemini API (we call to the Gemini API).
Interesting. We don't want to bombard the UI with noise though. But is this a type of banner you'd want to see?
Maybe 🤔 roadmap or thing coming in December or next year something too look forward 2. Gemini 3 great. I really love Jules agent think is by far the most easy going product and think it has so much going for it. @formal tusk maybe connect every session to your google Drive or something or form ur some smart way connect google different echo system that would be something. Are we talking 2 weeks before Gemini 3 comes back?
maybe more friendly.. just a small 'context used' or something, so we have a metric to go by
Thanks for the update. If nothing else changes, maybe the issues I am seeing will resolve when it gets switched back to 3.
Think should be metrics that will refine the task and start new task with info so it does not get locked
Or divide this multiple promt or fork this and have connector hub for different task refine or update aspects to this project, or option delete files create if they are totally garbage 🗑️ Jules can sometimes do some bad things and better delete task all together or old task . What is take for manage this @formal tusk what is possible to do?
Thinks there is better way for solutions workflow near its limit start a new task or or refine data clear previous Data in session/chat. Think maybe should be called stages or workflow pyramids so it is easy structure key aspects and what focus should be and reference Doc.
Since the downgrade to Gemini 2.5, I have to correct Jules 9-10 times for tasks that previously took 1-2. The difference in context understanding is obvious. 🙁
Hey @polar swallow Thanks for your feedback, it clearly highlights the drop in efficiency. Please note that this is temporary and our team is working to restore it ASAP. Sorry for the inconvenience caused.
