Home / Errors / Bolt.new / prompt-is-too-long

Bolt.new 'Prompt is too long': hitting the context limit on larger projects

Bolt.new Moderate 20-45 minutes Last verified Jul 17, 2026
The fix  Shrink what Bolt loads into the model: add a .bolt/ignore file excluding tests, docs, and generated assets, delete dead code, split files over roughly 500 lines, and scope prompts to specific files. Start a fresh chat when responses turn inconsistent; the accumulated conversation is part of the overflow, and quality degrades well before the hard error appears.

Symptoms

  • A 'Prompt too long' style message on a project past the prototype stage
  • Incomplete or cut-off responses
  • Bolt forgets architectural decisions made earlier in the chat
  • Previously fixed bugs come back as regressions when Bolt makes new changes

Why it happens

Project files and chat history share one context window

Bolt feeds your project files plus the conversation history into the model together. As both grow, the combined payload exceeds what the model can process; community documentation cites limits on the order of 200K tokens free and 500K paid.

Truncation starts before the error does

Bolt truncates or loses earlier context as the window fills, so the model works from an incomplete picture of the code. Quality degrades, decisions get forgotten, and regressions appear before you ever see the hard error.

How to fix it

  1. Create a .bolt/ignore fileExclude tests, docs, and generated assets from Bolt's context. Everything the model does not load is context you get back for actual code.
  2. Delete dead code and split monster filesRun npx knip to find unused files and exports, delete them, and split any file over roughly 500 lines into focused modules. Smaller files mean less context consumed per prompt.
  3. Scope prompts and use Discussion mode for questionsAsk about specific files or functions instead of 'the app', and use Discussion or Plan mode for analysis so the full build context is not loaded just to answer a question.
  4. Start a fresh chat when quality degradesThe accumulated conversation is part of the problem, so start a fresh chat (or duplicate the project) when responses become inconsistent. Inconsistency is your early warning; the hard error is the late one.
  5. Re-state key constraints in new promptsDo not assume Bolt remembers earlier decisions. Restate them explicitly, for example: 'we use Supabase auth, do not touch the auth flow'.
  6. Know when you have outgrown the browserFor genuinely large apps, export to GitHub and continue in a local IDE agent. Browser-context tools degrade with codebase size; this is a structural limit, not something you can prompt your way around.

How to prevent it

  • Add a .bolt/ignore file in the project's first hour, not after the first 'Prompt too long'.
  • Keep files under roughly 500 lines; split them before they sprawl.
  • Run npx knip periodically to clear dead code.
  • Start fresh chats at natural milestones instead of one endless conversation.
  • Restate standing constraints in prompts rather than relying on the model's memory.

Frequently asked questions

What causes the 'Prompt is too long' error in bolt.new?

Bolt feeds your project files plus the chat history into the model's context window, and on a project that has grown past the prototype stage that combined payload exceeds what the model can process. Community documentation cites limits on the order of 200K tokens on the free tier and 500K on paid. The fix is shrinking what gets loaded: a .bolt/ignore file, dead-code deletion, smaller files, and fresh chats.

Why does Bolt forget decisions and reintroduce old bugs?

When the combined project-plus-chat context outgrows the model's window, Bolt truncates, and earlier decisions and fixes fall out of the picture. The model then works from an incomplete view of the code, which is why previously fixed bugs come back as regressions. Restate key constraints in new prompts and start fresh chats instead of relying on Bolt's memory.

What is bolt.new's context limit?

Community documentation cites limits on the order of 200K tokens for free accounts and 500K for paid ones. The practical limit is lower because project files and chat history share the same window, and quality degrades before the hard error appears. Treat inconsistent responses as the real signal that you are near the ceiling.

Should I move my project off bolt.new when it gets big?

For genuinely large apps, yes: export the project to GitHub and continue in a local IDE agent, because browser-context tools degrade with codebase size. Before that point, a .bolt/ignore file, dead-code deletion, files under roughly 500 lines, and tightly scoped prompts buy you significant headroom.

Sources

Related errors

Cursor Forgot Everything Mid-Chat (Context Loss in Long Sessions)

Claude Code Forgot Everything After Auto-Compact

Bolt.new wasting tokens: failed fixes and whole-file rewrites draining your balance

Replit Agent broke my code: features it wasn't asked to touch

The Vibe Oops briefing

One email when something ships to production that should not have. New incidents, new error guides, no filler.