Where does the bug appear (feature/product)?
Somewhere else…
Describe the Bug
Problem Description:
I’ve noticed that when interacting with the Chat assistant on the Cursor documentation site (or within the IDE integration), the connection frequently terminates or cuts off abruptly if the response generation exceeds 60 seconds.
Technical Observation:
- It seems there is a hard timeout limit at the 60s mark, which is often insufficient for complex reasoning tasks or when processing large context files (rules/docs).
- For reference, high-performance LLM deployments on platforms like Vercel typically recommend or support execution limits between 500-800 seconds to account for high-latency model streaming and deep-thinking processes.
Impact:
- Users lose the end of the response, requiring manual retries which consume quota and break the development flow.
- Complex “Rules” or “Context” analysis often hits this wall because the model takes longer to synthesize multiple documentation sources.
Suggested Fix:
Please consider increasing the maxDuration or gateway timeout settings for the Chat endpoint to align with industry standards (e.g., 300s+) to ensure stability for long-form generations.
Steps to Reproduce
- Environment Setup: Open the Cursor official documentation (e.g.,
cursor.com/docs/context/rules) and initiate a conversation with the built-in Chat assistant. - Trigger Complex Task: Provide a prompt that requires heavy context processing or long-form code generation.
- Example: “Analyze all the rules in the current documentation and generate a comprehensive
.cursorrulesfile for a complex Monorepo project with 20+ microservices.”
- Monitor Execution Time: Observe the streaming response. The model begins generating correctly, but the connection remains active for an extended period.
- Observe the Interruption: Exactly around the 60-second mark, the streaming stops abruptly.
- Verify Network Error (Optional but recommended): Open Browser DevTools (F12) → Network Tab. Observe a 504 Gateway Timeout or a fetch stream cancellation from the backend API.
Expected Result
The assistant should continue streaming the response until the generation is complete, ideally supporting a duration of 500-800 seconds (consistent with high-performance Vercel/Edge deployment standards) to accommodate complex reasoning.
Actual Result
The connection is forcibly closed by the server/gateway at 60 seconds, resulting in an incomplete response and a broken user experience.
Expected Behavior
- Continuous Stream Duration: The API connection should remain open for the entire duration of the model’s reasoning and generation process. Given the complexity of analyzing context rules, the
maxDurationfor the chat function should be extended beyond the default 60-second threshold. - Alignment with Industry Standards: To ensure a stable developer experience, the timeout limit should align with high-performance Vercel/Edge deployment recommendations (500-800 seconds), allowing for deep-thinking tasks and large-scale code synthesis.
- Graceful Error Handling: If a legitimate system timeout occurs, the UI should provide a clear “Resume” or “Retry from last chunk” option, rather than abruptly terminating the stream with an incomplete response.
- Context Integrity: The assistant should be able to finish generating comprehensive
.cursorrulesor multi-file documentation summaries without being cut off by gateway-level constraints.
Operating System
Windows 10/11
Version Information
user in chrome
For AI issues: which model did you use?
all model have same issues
Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor