[Bug]: Agent runs terminate at exactly 5 minutes due to undici body/header timeout in @librechat/agents

[Learn-Implement]

What happened?

Long-running Agent requests terminate at exactly 5 minutes with:

error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type terminated

This occurs despite all of the following:

• HTTP_REQUEST_TIMEOUT_MS=900000 (15 minutes) set in .env and confirmed present inside the API container via docker-compose exec api printenv HTTP_REQUEST_TIMEOUT_MS.
• waitForRun timeout in api/server/services/Runs/handle.js patched to 60000 * 15 and confirmed active inside the container via grep.
• Redis enabled with USE_REDIS=true and USE_REDIS_STREAMS=true, confirmed PONG from redis-cli.

Expected: Agent runs should respect the configured 15-minute timeout and complete normally.

Actual: Agent runs are terminated at exactly 5 minutes. The 5-minute mark matches undici's default bodyTimeout and headersTimeout (300,000 ms).

Precedent: This exact issue was already fixed for MCP connections in PR #9476, where an undici Agent with bodyTimeout: 0 and headersTimeout: 0 was configured to prevent the 5-minute cutoff. This report requests the same fix be applied to the OpenAI/DeepSeek client used by the Agent controller.

Presets are less affected than Agents under identical long-running queries, suggesting the Agent networking path is the source of the cutoff.

Version Information

LibreChat v0.8.7
Deployment: Self-hosted Docker Compose (local)
Image: registry.librechat.ai/danny-avila/librechat-dev:latest
Reverse proxy: None (direct port 3080)
Custom endpoints: DeepSeek V4 Pro, Gemini, Qwen
Redis: redis:7-alpine (docker), USE_REDIS_STREAMS=true

Steps to Reproduce

1. Configure an Agent using the DeepSeek endpoint (deepseek-v4-pro) with reasoning_effort: high.

2. Set HTTP_REQUEST_TIMEOUT_MS=900000 in .env and pass it to the api service in docker-compose.override.yml.

3. Patch api/server/services/Runs/handle.js: change timeout = 60000 * 3 to timeout = 60000 * 15 and mount it via docker-compose.override.yml. Confirm inside the container with:
   docker-compose exec api grep -n "timeout = 60000" /app/api/server/services/Runs/handle.js

4. Enable Redis with USE_REDIS=true and USE_REDIS_STREAMS=true.

5. Send a complex query that requires more than 5 minutes of high-effort reasoning.

6. Observe the API logs. At exactly 5 minutes, the request terminates with:
   error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type terminated

What browsers are you seeing the problem on?

Microsoft Edge, Firefox, Chrome

Relevant log output

2026-09-13 17:40:34 error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type terminated
2026-09-13 17:44:51 error: [api/server/controllers/agents/client.js #sendCompletion] Unhandled error type terminated

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct