search

stackblitz / bolt.new

Prompt, run, edit, and deploy full-stack web applications
https://bolt.new
MIT License
8.73k stars 3.42k forks source link

Chat Error: prompt is too long: 200000 tokens > 199999 maximum #1322

Closed kc0tlh closed 6 days ago

kc0tlh commented 3 weeks ago

📣 Issue Resolved: this fix has been merged into production and bolt.new now supports projects up to 2.5x larger!

Thank you to everyone in the thread above providing workarounds and helping each other. This is MUCH MUCH appreciated, as we are currently in the process of training our newly hired support staff, and working out of a backlog on the support@stackblitz.com/hello@stackblitz.com email inbox (we will respond to EVERY SINGLE request).

If you are still looking for support for other issues, please start with our newly launched support center 🎉

What does this error message mean? This error message means that your project has become too large for Bolt to sync the entire project with the AI.

As described in issue #678, Bolt.new syncs both the chat history & file system with the AI's memory. When this error occurs, it means that the combination of message history and file system has become too large for the AI model to work on under the current configuration.

What to do when you get this error Here are a couple approaches to consider:

What are we doing about this? We are currently in the R&D phase to identify application-level changes we can make to better manage the AI context window. Our current approach has the benefit of making Bolt.new very simple to use, as the AI is constantly aware of the entire context of your application so you don't have to tell it which files to focus on. However, it does introduce issues like the context window limitation described above. Stay tuned to this issue for future updates!

Welexon commented 3 weeks ago

same issue for me, over and over again. prompt is too long: 201843 tokens > 200000 maximum. i have not received any refund and bought 30 mill tokens, now to test again. still same issue. I even tried forking. I kindly ask for a refund, my entire project has halted. I sent an email about refund today.

user: Welexon https://bolt.new/~/sb1-g7m98y

behnod24 commented 3 weeks ago

Hello, I get the same error, could you please fix this issue? I could not do anything because of this issue.

Thank you for your efforts and supports.

tuanvuvo commented 3 weeks ago

same as me. can not work when just secord request blocked of this

heliusmaximus08 commented 3 weeks ago

same issue. this is really sad

paydayrich commented 3 weeks ago

@kc0tlh how can i request a refund, I added tokens to have the the chat go into endless loop of errors and now I am having the same issue of the prompt being too long. my project has been stuck for 3 days.

danthemank commented 2 weeks ago

I used an approach with an automation where I brought the context for a desired outcome in stages where a thing was being affected by a portion of the total context, then again by another portion of the context, through finally the remainder of the context before considering the "affect" to have been completed. Maybe there's merit to an approach like this here?

Either way, I am kind of stuck on a project I was working on given this problem and will try the ignore method next as clearing other elements from the context window or forking did not work.

Until the context window issue is solved, Bolt may very well be the most amazing prototyping tool of all time, but through limitations of the models themselves (which are not theirs, but Anthropics), will not be able to continue on after a certain point.

For Bolt: I think the tech is amazing and I've upgraded my paid account a few times to keep going out of excitement. I think some of the frustration seen here in the Issues section could be solved by some blatant warnings and clarifying the marketing to explain current limitations. Also, not understanding the context window as someone just getting started meant I went through like 200k tokens at a time with some of my messages which was astonishing, but I accepted because the results were still WAY MORE than I could have paid the same money for to myself for my time or any other human developer.

To anyone else, these people are trying to make a great tool. I'm sure you can imagine they don't want problems. Don't be a jerk.

Habartru commented 2 weeks ago

This is a very strange behavior of BOLT developers! First of all, you do not warn that at one point the system may stop working! Secondly, you have not entered any indicator to understand how many tokens are already in the context window! Thirdly, you didn't even bother to make some kind of cleaning tool. Fourth, and most importantly! Users who are able to achieve maximum context should be saved as much as possible, because in my case, I have already brought you $ 450 this month, and you are not doing any of the above! And fifth, since there is such a limitation of the context window for the model, then introduce an additional agent, otherwise people will not do anything serious on your platform and will just run away from you, and you don't seem to have started badly!!!!! And I would also like to say that support is very important and the fact that you do not respond to requests and people waste a lot of time waiting for some kind of reference from you is not right. So you spit in their face... It's a shame, of course. You could get better, but apparently you don't need it!

Frankg40 commented 2 weeks ago

What's unacceptable is not telling future users there will be a project size limitation.

People are spending money on tokens nonstop to continue building out their project unknowingly that they will reach a point they can no longer continue.

You have to put a warning out immediately or find a more efficient way to provide context.

I've spent a few hundred dollars on building a comprehensive app only to find out I cant build a comprehensive app.

danthemank commented 2 weeks ago

I hear ya and I did the same @Frankg40. Only saying a few hundred dollars for something that otherwise would cost tens of thousands or more... communication by Bolt should be better, but they should get a little grace for this. The value I have been able to get out of the use of this system in a little over 24 hours FAR outweighed the expense for what building without the tool would have cost.

I built a near complete SaaS MVP in a day within a couple hundred bucks. Right now I can't even deploy due to context window and in fact there are errors that can't be resolved by Bolt for the same reason, but even if I have to pick it up from here to fix issues and continue building it was still amazing value.

danthemank commented 2 weeks ago

@kc0tlh have you guys looked into Claude Enterprise versus Sonnet 3.5? 500k token context window instead of the 200k, just not sure how different the model will perform for this application.

heliusmaximus08 commented 2 weeks ago

@danthemank what is claude enterprises? something like bolt.new? are you thinking of import the code from bolt into claude?

Frankg40 commented 2 weeks ago

Increasing token size limit is not the answer. 200k tokens per prompt is already not a sustainable business model.for something that gets it wrong more than right.

danthemank commented 2 weeks ago

Increasing token size limit is not the answer. 200k tokens per prompt is already not a sustainable business model.for something that gets it wrong more than right.

@Frankg40 I do very much agree with you there on the sustainability of the tokens per prompt. I would say my experience was honestly not to have more problems than efficient quality outcomes though and for the record, I'm not paid nor do I have any affiliation with Stackblitz / Bolt.

danthemank commented 2 weeks ago

@danthemank what is claude enterprises? something like bolt.new? are you thinking of import the code from bolt into claude?

@heliusmaximus08 I am referring to the Enterprise model offering from Claude where it was mentioned above that Sonnet 3.5 from Claude is currently used in Bolt. Not sure what the implications of changing one for the other functionally would have, but the one main difference in the two is the size of the context window (and the costs).

Frankg40 commented 2 weeks ago

For those of us interested in a fix for this even if we cant fix it ourselves. I just had a convo with Bolt.

RIght now you are costing almost 200,000 tokens per prompt. Break down the token sizes when I prompt you for my project right now

Let me break down the approximate token usage in your current project:

Core Project Files (~25,000 tokens): React components TypeScript interfaces Utility functions Store/state management Hooks Configuration Files (~5,000 tokens): Package.json TypeScript configs ESLint configs Vite config Firebase configs Database/Backend Files (~8,000 tokens): Firebase rules Database schemas Migration scripts Data transformers Admin/Management Files (~7,000 tokens): Admin components Mapping interfaces Field management Database selectors Onboarding/Auth Files (~5,000 tokens): Authentication guards Onboarding steps User management System Instructions/Constraints (~3,000 tokens): WebContainer limitations Technology preferences Response guidelines File Tracking/Metadata (~2,000 tokens): File modifications Running commands Hidden files info Total: ~55,000 tokens per interaction

This is significantly more efficient than the 200,000 tokens you mentioned, but there's still room for optimization by:

Splitting large components into smaller ones Moving utility functions to separate modules Optimizing database schemas Reducing duplicate code Streamlining configuration files Would you like me to help optimize any specific area to reduce the token count?

Ok now give me the token amounts for my project when i prompt you for the following

_File modifications and changes Currently running commands Project configuration Development environment details

Here's the breakdown of token usage when I receive your prompts:

File Modifications (~150,000 tokens): Complete file contents for all modified files File diffs showing changes Removed files tracking Hidden files list

Running Commands (~500 tokens): Current process information Command status Runtime details

Project Configuration (~15,000 tokens): Package.json contents Environment settings Build configurations Dependencies list TypeScript/ESLint configs

Development Environment (~5,000 tokens): WebContainer specs Available commands Runtime limitations Technology preferences System constraints

Total: ~170,500 tokens per interaction

The largest portion comes from file modifications since I receive complete file contents rather than just diffs. This ensures I have accurate context but does increase token usage significantly._

Does this mean if I want to change the background color on the newsfeed, Bolt is going to send AI my entire project even if 95% of it is irrelevent for the change I need to make?

_Yes, that's correct. Even for a simple change like modifying the background color in the NewsFeed component, the entire project context is sent with each interaction. This includes all file contents, configurations, and environment details.

While this ensures I have complete context and can make informed decisions, it's not very efficient for small, isolated changes where only a specific component needs modification.

Unfortunately, there's currently no mechanism for me to receive only the relevant files or partial context. Each interaction includes the full project state, even if the change only affects a single line of code._

Im spending probably 8-12 hours a day on Bolt so Ive learned its quirks pretty well.

The only thing its also not including in the token count is in the chat container. Ive found clearing the chat container OFTEN after successfuly fixing anything where I dont need to rollback helps in reducing token size and possibly even more importantly it makes the AI more focused on what you want and does a better job coding.

The second thing that SIGNIFICANTLY increases the efficiency and lessens errors is when any file or page youre working on get more than 200 lines of code, I ask it tif it would be a good idea to modularize the page or function. It often agrees. If you don't follow... Say you have Facebook. You have the navigation at the top, then you have status update, then you have reels, then you have the newsfeed.

This would take the newsfeed page and put each function in separate files. So it would have:

HomePage Sub file > Nav Sub File > Status Update Sub FIle > Reels Sub File > Newsfeed

Other than this being good coding conduct in general, specifcally for bolt it is crucial. Because Bolt likes to rewrite your ENTIRE page. And when it does it often forgets to put things back, or loses context altogether and you get a completely different looking page. Doing it like this, when you make a change to say REELS, instead of it rewriting the entire page, its just going to change the REELS file. It reduces risk dramatically for errors or large unrequested changes.

I asked Bolt

Does modularizing pages and components increase token size a lot or does it just split the token into smaller pieces?

_Modularizing code into separate components and files doesn't increase the total token count significantly - it primarily redistributes the same tokens across multiple files. The actual code content remains roughly the same, just organized more efficiently.

For example, if you have a 1000-token monolithic component and split it into 5 modular components of 200 tokens each, you're still dealing with approximately the same total token count. The small increase comes from:

Additional import/export statements Component declarations and props interfaces File headers and basic setup code However, these minor increases are negligible compared to the benefits:

Better code organization Easier maintenance Improved reusability Better performance through code splitting Simpler testing Clearer separation of concerns The token count increase is typically less than 5-10% when properly modularizing code, which is a worthwhile tradeoff for the significant improvements in code quality and maintainability._

Couple LARGE points to make here.

To Bolt -NEW users need to be made aware of the current platforms Ceiling. People are not aware that when a project reaches a certain size, Bolt can no longer handle it. When people don't know that, they believe they can build whatever they want. They spend $500 on Tokens building a comprehensive app, only to find out they cant build a comprehensive app because it was not said up front...

ALSO - That as your app gets bigger more and more tokens will be used every time you interact with it. At first it was weird that it seemed like I was going through tokens WAY faster as I built out the app then when I started. That's because we are. Increasing token size is not the fix. Or people will get to a point with their app where theyre spending 1M tokens per prompt, which even then might be fine if it worked perfectly. But you cant be dropping a Million on tokens to get errors.

I know you are working on a fix, but users should still be made aware.

Frankg40 commented 2 weeks ago

**You said my project size is roughly 55k tokens.

If I were to reduce it to 45k tokens.

What would the total token reduction be when prompting for a file modification?**

Let me break this down:

Current scenario (55k project):

File modifications: ~150,000 tokens per prompt Total with context: ~200,000 tokens Reduced scenario (45k project):

File modifications: ~122,000 tokens per prompt Total with context: ~164,000 tokens Total reduction per prompt: ~36,000 tokens

This reduction would come from:

Smaller file sizes More efficient component structure Less redundant code Better code splitting The key point is that while we're only reducing the project size by 10k tokens, the savings are multiplied in each prompt because of how file modifications and context are tracked. Each token saved in the base project results in approximately 3.6 tokens saved per prompt due to the tracking of modifications, diffs, and context.

So basically. if your project size is 5k tokens. Its going to cost roughly 15k tokens directly based on the amount of files to really complete a modification.

When I first asked it to break down the token amounts because I was using 200k token per prompt, it told me my project was only 55k. So I figured reducing project size wasnt going to do much, but it does.. because its essentially tripling your project size in tokens every interaction.

Hope this helps.

Frankg40 commented 2 weeks ago

Also going to share this year to try and help you guys even though Im just a regular user:

**please review all of my files for

  1. deprecated or unused code or files that have been left that can be removed. Give me a list of and token amount for each. 2. How many tokens would removing the entire admin - database management section? Answer only**

Deprecated/Unused Files & Code:

Files that can be removed:

  • src/views/PostStylesShowcase.tsx (4,892 tokens)
  • src/views/PostStylesShowcaseV2.tsx (3,756 tokens)
  • src/components/AddLogModal.tsx (2,143 tokens)
  • src/components/VoiceInput.tsx (1,245 tokens)
  • scripts/showHeaders.mjs (642 tokens)
  • scripts/showAllHeaders.mjs (1,324 tokens)
  • scripts/findMaxLengths.mjs (2,156 tokens)
  • src/components/SocialInteractions.tsx (1,876 tokens)
  • src/components/food/ShareDiaryConfirmation.tsx (2,234 tokens)

Total tokens that can be removed: 20,268 Admin/Database Management Section:

Removing the entire admin section would reduce:

  • src/views/admin/* (4,892 tokens)
  • src/components/admin/* (8,756 tokens)
  • src/lib/admin/* (6,234 tokens)
  • src/hooks/admin/* (2,143 tokens)
  • src/lib/mapping/* (7,654 tokens)
  • scripts/utils/transformers/* (3,245 tokens)
  • Related types and interfaces (1,245 tokens)

Total tokens that would be removed: 34,169

I created an admin section that really shouldnt be part of my app for launch anyway. I downloaded the whole project on StackBlitz so I still have the admin section files. But now I can remove the admin section, and the deprecated files, which will be a large token reduction for the project and per prompt.

danthemank commented 2 weeks ago

Thanks for sharing @Frankg40 - based on what you said, using the ignore method would likely be even more useful if the level of modularity was much higher as well.

digikordeveloper commented 2 weeks ago

Please, guys, I paid almost $500 for this project, and for the last 4 days, I can't continue to build my app. This is urgent. I see that you guys are working on all types of projects and features, and that is very nice. But all of your power needs to be on this problem. I have a deadline; I can't just wait with this problem. Please help urgently!!!

salimmallick commented 2 weeks ago

@digikordeveloper IF you got stuck suddenly after progressing a lot, try to restore from a slightly previous backup and strategize with bolz on how much tokens are being consumes and how you should proceed further without being stuck in the situation. Try this approach and see if it helps. At least you will be able to move forward.

digikordeveloper commented 2 weeks ago

@salimmallick thank you for your answer but that won't help me. I need to be able to work with nothing like this happening again. I really hope bolt will fix this asap.

Frankg40 commented 2 weeks ago

@salimmallick thank you for your answer but that won't help me. I need to be able to work with nothing like this happening again. I really hope bolt will fix this asap.

You know this is a beta right? Maybe don't rely on a brand new platform in beta for a project with a deadline.

digikordeveloper commented 2 weeks ago

@Frankg40 good point. But they also should not allow to buy $500 worth of tokens just to get stuck with the project.

Trisfidel commented 2 weeks ago

@salimmallick thank you for your answer but that won't help me. I need to be able to work with nothing like this happening again. I really hope bolt will fix this asap.

You know this is a beta right? Maybe don't rely on a brand new platform in beta for a project with a deadline.

@Frankg40: If you don't get to Bolt.new via GitHub, there is NO mention of it being a beta product in the customer journey as far as I could tell. I ran into this same trap with Bolt. I re-checked, but if you go straight to the Bolt.new site and then subscribe after using the free version, there's no mentioning of beta, nor its current limitations.

Trisfidel commented 2 weeks ago

@Stackblitz,

As a resident of the EU, I believe your current approach to selling a beta product may conflict with European consumer protection laws. Specifically:

Lack of Transparency: EU law mandates informed consent before purchase. The sales process on the Bolt.new website does not even mention that this is a beta product, let alone disclose its limitations. This lack of transparency could be seen as misleading.

Refund Policy: Under EU regulations, consumers are entitled to a refund if the product is non-functional or if it does not match the description provided. I was unable to find any option for requesting a refund on your platform.

Given these concerns, I am now left with no choice but to escalate the matter through my credit card provider to resolve this issue and recover my funds.

This experience is both disappointing and frustrating. I had high hopes for completing my project with Bolt, as it is an impressive and promising software. However, these legal and customer service shortcomings have unfortunately marred my experience.

Additionally, I am encountering the error message: "prompt is too long: 200612 tokens > 200000 maximum." This issue has now locked me out of interacting with the AI, even outside of coding tasks, and your service does not provide an option to fix this known issue.

Please address these matters promptly to ensure compliance and improve trust with your EU-based users.

kc0tlh commented 2 weeks ago

Thank you all for sharing such valuable feedback. We deeply appreciate the StackBlitz community’s involvement in helping us improve Bolt.new. We understand how frustrating it can be to encounter roadblocks, especially after investing significant time and resources into your projects.

As a small team of 20 who launched Bolt.new just over a month ago, we’re actively working to address the AI context window limitations, optimize token usage, and improve communication. Here’s a breakdown of our responses to key issues raised:

  1. Context Window Limitations

    • The "prompt is too long" error occurs because the AI model (Anthropic Sonnet 3.5) has a maximum 200,000-token context limit, as noted above. We recognize this impacts workflows for many of you.
    • To mitigate this, we’re exploring application-level optimizations to streamline token usage while maintaining the AI’s understanding of your project’s context. This is a complex task requiring ongoing R&D.

  2. Token Efficiency and User Tips

    • @danthemank and @Frankg40 shared useful tips on modularizing code to reduce token strain—thank you! Breaking up files can help manage token usage and limit context overflow.
    • Advanced users may also use the .bolt/ignore file to exclude non-essential files from the AI context. This can free up token space, though it may affect how the AI interprets your project, so use it thoughtfully.
    • Clearing chat history regularly can further help manage token usage by removing unnecessary data from the context window. We’re also reviewing suggestions to add a token usage indicator for more control over token consumption.

  3. Refunds and Customer Support

    • @Welexon, @paydayrich, and others: Refunds are available to users impacted by these limitations, and our team is working through requests as quickly as possible. For assistance, please reach out to hello@stackblitz.com.
    • @Trisfidel, we hear your concerns about EU consumer protection laws. We’ll be updating the Bolt app to make the refund process clearer for all users.

  4. Transparency on Limitations

    • We hear you, @Frankg40, @Trisfidel, and @Habartru, on the importance of transparency. We will be updating in-app messaging to clarify the context window limitation and token consumption details, aiming to make these limitations more visible and avoid future frustrations.

To everyone who has contributed to this discussion, thank you for your patience, understanding, and insights. We’re fully committed to addressing these issues and making Bolt.new better based on your feedback.

Thank you for helping us shape Bolt.new into a tool that better serves you!

mkaufmanentrep commented 2 weeks ago

@salimmallick - Thats so good to know, I need to implement this before the Bolt team gets around this. Anyway you could make some documentation or a loom video that would show how you did this. I know its an ask but just a shot

salimmallick commented 2 weeks ago

@mkaufmanentrep You can have a conversation with bolt regarding this and bolt will help you optimize the code base so that you can minimize the token consumption and reduce the prompt length. It will even create the needed documentation around it

mkaufmanentrep commented 2 weeks ago

@mkaufmanentrep You can have a conversation with bolt regarding this and bolt will help you optimize the code base so that you can minimize the token consumption and reduce the prompt length. It will even create the needed documentation around it

That would be wonderful, unfortunately I'm not even able to have a conversation with the base as i'm very over limits on the project. I'm a rookie

salimmallick commented 2 weeks ago

@mkaufmanentrep i have been in that situation myself. The one work around I figured out was to go back to a previous version from the backup which is available in the settings part and then restore a backup. Then you can have a conversation and strategize with bolt on which components can be disabled and hidden till you complete other elements of the page.

gboross commented 2 weeks ago

Hi everyone, could you please help me understand how to use the .bolt/ignore file? Where should I create the file, how do I load it, and what format should I use to write in it exactly?

Frankg40 commented 2 weeks ago

Hi everyone, could you please help me understand how to use the .bolt/ignore file? Where should I create the file, how do I load it, and what format should I use to write in it exactly?

Yeah, same... not sure how to use this.

danthemank commented 2 weeks ago
  • ADVANCED USERS ONLY: .bolt/ignore In every bolt project, if you open it in StackBlitz you can edit a file called .bolt/ignore, and in this file you can list out any folders or folders that should be excluded from the AI context window. For example, here is our vite react starter’s ignore files: https://stackblitz.com/edit/vite-shadcn?file=.bolt%2Fignore. Any files listed there will be completely invisible to the AI, and will clear up space in the context window. You’ll need to edit the .bolt/ignore file in StackBlitz and then reopen the project in bolt for the changes to take effect. Please note: hiding files from the AI can have unintended consequences as it is no longer aware of your entire project. This approach is very powerful, but is only recommended for advanced users who can make informed decisions about what can safely be excluded, and understand/resolve issues that may arise from this approach.

If the ignore file does not exist create it. I named it "ignore" and that's it. I added each folder / subfolder on a separate line. I used to signfify "all", example: folder/subfolder/ (everything inside of the subfolder). Nothing will tell you it is ignoring anything explicitely, you'lll just suddenly have a smaller token window which is cool, but you do need to be very aware of what you've hidden from the AI's ability to see and use while continuing to work on the project.

gboross commented 2 weeks ago
  • ADVANCED USERS ONLY: .bolt/ignore In every bolt project, if you open it in StackBlitz you can edit a file called .bolt/ignore, and in this file you can list out any folders or folders that should be excluded from the AI context window. For example, here is our vite react starter’s ignore files: https://stackblitz.com/edit/vite-shadcn?file=.bolt%2Fignore. Any files listed there will be completely invisible to the AI, and will clear up space in the context window. You’ll need to edit the .bolt/ignore file in StackBlitz and then reopen the project in bolt for the changes to take effect. Please note: hiding files from the AI can have unintended consequences as it is no longer aware of your entire project. This approach is very powerful, but is only recommended for advanced users who can make informed decisions about what can safely be excluded, and understand/resolve issues that may arise from this approach.

super, thanks.

Frankg40 commented 2 weeks ago

Even better guys. here's what you do.

First - Reach a point in your project where you do not need to roll anything back (You will still have your backups). If you had to close the project or refresh the browser right where you are at, it would be fine. The change you are going to make is not going to ruin or delete anything. I'm only suggesting this just because it makes the most sense that you do this when you currently don't need to roll back.

I prompted Bolt:

ME: I plan on working on just the diary page to make these x y z features to work. I want to use the .bolt ignore feature.

Give me a list of all my files that the AI does not need to see for me to work on this specific part of my app.

Bolt then made a text file (just for a record) and gave me a list of all the files with comments and categorized.

  1. Copy the list.
  2. Open the project in Stackblitz.
  3. Find the .bolt folder (it's the first one)
  4. Click the create a doc icon (across from the bolt folder there is a few icons on the same line, it's the first one)
  5. Paste the ignore files Bolt gave you in the newly created .bolt > ignore doc.
  6. Click save.
  7. Close Stackblitz.
  8. Go back to Bolt and refresh your web browser.

Done.

If you decide to work on a different part of your app:

  1. Go back into StackBlitz
  2. Delete all the ignore files in the ignore doc.
  3. Click Save.
  4. Close Stackblitz
  5. Go back to Bolt > Refresh Web browser.
  6. Prompt Bolt again on what you intend to do and to give you all the files it does not need again. Repeat the process.

(some of those extra steps like close stackblitz, refresh web browser, may or may not make it work better - but it's made it seamless for me.)

Cheers

Frankg40 commented 2 weeks ago

@kc0tlh My suggestion is working really well by having bolt give me the files it can ignore based on the "feature" i want to work on.

Maybe add a function where you tell bolt this and it uploads the ignore files to the .ignore folder just like it adds rules to the firestore rules.

THen add a second prompt where you tell it you want to work on a new feature - it Clears the .ignore file. You tell it what you want to work on, and it uploads a new list of files to ignore.

A 3rd catch - If you ever tell it to do something, and it sees that feature doesnt exist, it maybe asks or looks at the ignore file list to see if what you want to work on might be on the ignore list (I know it wont be able to actually read the file in the ignore list) but if you are working on a feature and it's calling a variable or another file that Claude doesnt see, it can know theres a possibility the user has this file on ignore, and to maybe remove it. (More advanced for you guys to add, maybe not IDK)

Users just need to be given the instructions of this feature..

Just an idea. Ive prob reduced my tokens by 70% by having bolt tell me which files.

danthemank commented 2 weeks ago

@kc0tlh My suggestion is working really well by having bolt give me the files it can ignore based on the "feature" i want to work on.

Maybe add a function where you tell bolt this and it uploads the ignore files to the .ignore folder just like it adds rules to the firestore rules.

THen add a second prompt where you tell it you want to work on a new feature - it Clears the .ignore file. You tell it what you want to work on, and it uploads a new list of files to ignore.

A 3rd catch - If you ever tell it to do something, and it sees that feature doesnt exist, it maybe asks or looks at the ignore file list to see if what you want to work on might be on the ignore list (I know it wont be able to actually read the file in the ignore list) but if you are working on a feature and it's calling a variable or another file that Claude doesnt see, it can know theres a possibility the user has this file on ignore, and to maybe remove it. (More advanced for you guys to add, maybe not IDK)

Users just need to be given the instructions of this feature..

Just an idea. Ive prob reduced my tokens by 70% by having bolt tell me which files.

Really smart. One note, the AI having access to the ignore file would be (if it doesn't already) a pretty small number of tokens itself. I would think if it were aware of the total project filesystem, then had the list of things it doesn't have access to the codebase for, it should be able to make some reasonable assertions.

Frankg40 commented 2 weeks ago

@kc0tlh My suggestion is working really well by having bolt give me the files it can ignore based on the "feature" i want to work on. Maybe add a function where you tell bolt this and it uploads the ignore files to the .ignore folder just like it adds rules to the firestore rules. THen add a second prompt where you tell it you want to work on a new feature - it Clears the .ignore file. You tell it what you want to work on, and it uploads a new list of files to ignore. A 3rd catch - If you ever tell it to do something, and it sees that feature doesnt exist, it maybe asks or looks at the ignore file list to see if what you want to work on might be on the ignore list (I know it wont be able to actually read the file in the ignore list) but if you are working on a feature and it's calling a variable or another file that Claude doesnt see, it can know theres a possibility the user has this file on ignore, and to maybe remove it. (More advanced for you guys to add, maybe not IDK) Users just need to be given the instructions of this feature.. Just an idea. Ive prob reduced my tokens by 70% by having bolt tell me which files.

Really smart. One note, the AI having access to the ignore file would be (if it doesn't already) a pretty small number of tokens itself. I would think if it were aware of the total project filesystem, then had the list of things it doesn't have access to the codebase for, it should be able to make some reasonable assertions.

Thanks. You try it out or anything else? So far so good.

LacourQuentin commented 2 weeks ago

@kc0tlh My suggestion is working really well by having bolt give me the files it can ignore based on the "feature" i want to work on.

Maybe add a function where you tell bolt this and it uploads the ignore files to the .ignore folder just like it adds rules to the firestore rules.

THen add a second prompt where you tell it you want to work on a new feature - it Clears the .ignore file. You tell it what you want to work on, and it uploads a new list of files to ignore.

A 3rd catch - If you ever tell it to do something, and it sees that feature doesnt exist, it maybe asks or looks at the ignore file list to see if what you want to work on might be on the ignore list (I know it wont be able to actually read the file in the ignore list) but if you are working on a feature and it's calling a variable or another file that Claude doesnt see, it can know theres a possibility the user has this file on ignore, and to maybe remove it. (More advanced for you guys to add, maybe not IDK)

Users just need to be given the instructions of this feature..

Just an idea. Ive prob reduced my tokens by 70% by having bolt tell me which files.

Thanks for the solution. It works well for me at first but still got the "prompt is too long" error at the end. Hope the bolt team will be able to find a solution soon.

danthemank commented 2 weeks ago

@kc0tlh My suggestion is working really well by having bolt give me the files it can ignore based on the "feature" i want to work on. Maybe add a function where you tell bolt this and it uploads the ignore files to the .ignore folder just like it adds rules to the firestore rules. THen add a second prompt where you tell it you want to work on a new feature - it Clears the .ignore file. You tell it what you want to work on, and it uploads a new list of files to ignore. A 3rd catch - If you ever tell it to do something, and it sees that feature doesnt exist, it maybe asks or looks at the ignore file list to see if what you want to work on might be on the ignore list (I know it wont be able to actually read the file in the ignore list) but if you are working on a feature and it's calling a variable or another file that Claude doesnt see, it can know theres a possibility the user has this file on ignore, and to maybe remove it. (More advanced for you guys to add, maybe not IDK) Users just need to be given the instructions of this feature.. Just an idea. Ive prob reduced my tokens by 70% by having bolt tell me which files.

Really smart. One note, the AI having access to the ignore file would be (if it doesn't already) a pretty small number of tokens itself. I would think if it were aware of the total project filesystem, then had the list of things it doesn't have access to the codebase for, it should be able to make some reasonable assertions.

Thanks. You try it out or anything else? So far so good.

Honetly the usefulness after a certain point of development diminishes. Unless you as the developer really know what is going on and the tech stack used. Even still, things will get updated and things that should get updated based on those changes will not (if they're in "hidden" files or the connection to a non-"hidden" but related file is "hidden"). Probably I will not be able to continue using bolt at all for the main project I am working on any longer, but I may try to spin up some other ideas to use up the rest of the credits at least. Bullish on the tech, but its place at least for now is for smaller projects or prototyping / MVP depending on size.

Frankg40 commented 2 weeks ago

@kc0tlh My suggestion is working really well by having bolt give me the files it can ignore based on the "feature" i want to work on. Maybe add a function where you tell bolt this and it uploads the ignore files to the .ignore folder just like it adds rules to the firestore rules. THen add a second prompt where you tell it you want to work on a new feature - it Clears the .ignore file. You tell it what you want to work on, and it uploads a new list of files to ignore. A 3rd catch - If you ever tell it to do something, and it sees that feature doesnt exist, it maybe asks or looks at the ignore file list to see if what you want to work on might be on the ignore list (I know it wont be able to actually read the file in the ignore list) but if you are working on a feature and it's calling a variable or another file that Claude doesnt see, it can know theres a possibility the user has this file on ignore, and to maybe remove it. (More advanced for you guys to add, maybe not IDK) Users just need to be given the instructions of this feature.. Just an idea. Ive prob reduced my tokens by 70% by having bolt tell me which files.

Thanks for the solution. It works well for me at first but still got the "prompt is too long" error at the end. Hope the bolt team will be able to find a solution soon.

You need to modularize your files.

Put 2/3 of your file on ignore. Ask it to review the 1/3 it can see and if any should be modularized. Repeat for the rest of the files.

The files not on ignore are still too big for the AI to receive.

heliusmaximus08 commented 2 weeks ago

OK thank you all guys. one ANOTHER Question: If i make a previous Backup to be under the Tokenlimit, do i lose the current backup file? How can i do this, not lose the current stadion of my app?

StackShipping commented 2 weeks ago

I'm wondering how I need to proceed now. I just upgrade to the 200 per month package and I cannot continue on my project. Should I request a chargeback on my credit card? Bolt, please fix this asap. I guess this should be that hard of a problem for you guys and when I am looking at the above complaints, I guess this deserves your priority! Will wait a maximum of 2 days before I will charge back my cc. I hope you guys manage to solve the issue!

I think a token crediting to all who suffered would be in place as a good gesture!

plataformaconcreta commented 2 weeks ago

Any news? I work almost everything in bolt, and now that my CRM is being completed I can't work. Please find a solution.

plataformaconcreta commented 2 weeks ago

prompt is too long: 201599 tokens > 200000 maximum

Again.....Please save my life!

kmlacr commented 2 weeks ago

Its boring come on guys fix that or give us refund back.

avxobro commented 2 weeks ago

99% finished in on my project and hit these limits. Found this thread. What a disaster. Literally was SO pumped to push my app into SEO history! FML.

wundulla commented 2 weeks ago

i have 65m tokens left of my 120mils i purchased. and no way to fix this token limit issue! Refund or a fix is required asap!

avxobro commented 2 weeks ago

Requested a refund until this app is passed beta. You can't sell a service with MILLIONS of tokens that can't be used to finish a project. Will resubscribe when it's debugged and worth the money. Also the tokens consumption rate is too high for the continual repairing what the AI broke which was already 100% fine.

danthemank commented 2 weeks ago

When in the history of development have you been able to create a complete and fully-featured application from the ground-up for a couple hundred dollars? The messaging on behalf of Bolt should be clearer, no doubt about that, but to say that what you're getting is not worth the money doesn't seem reasonable.

Accept a little responsibility for your own use of the tool and use the ignore method as mentioned above in various comments in the thread including at the very top of the page to continue working on your project.