Deep Research Timeout Issue

[xoChrisCo]

What happened?

I am having difficulties getting deep research to work. It returns with Error connecting to server, try refreshing the page.

Debugging:

• Double check that we have access to o3-deep-research and o4-mini-deep-research in https://platform.openai.com/settings/organization/limits
• Activate and deactivateDisable Streaming in the agent builder
• Test both o3-deep-research and o4-mini-deep-research with same results

Version Information

hcr.io/danny-avila/librechat-dev-api latest 8209b31dd7c8 4 hours ago 1.26GB
ghcr.io/danny-avila/librechat-rag-api-dev-lite latest 33b757cfadf6 41 hours ago 1.48GB

Steps to Reproduce

1. Create an agent "Deep Research".
2. Select model OpenAI -> 04-mini-deep-research
3. Enable Use Responses API and Web Search
4. Say "hi" to the agent. The agent responds "hi".
5. Ask the agent to do a deep research task
6. The agent is pending for 2-3 minutes and responds with Error connecting to server, try refreshing the page.
7. Refreshing the page just shows the agent reply with blank content.

What browsers are you seeing the problem on?

Chrome

Relevant log output

{"conversationId":"ebbb1476-b318-4319-a178-a3ccefbc7809","isCreatedByUser":true,"level":"debug","message":"[BaseClient] userMessage","messageId":"57539359-6a64-4596-9f06-98a9c1840b63","parentMessageId":"5c77d64d-72f1-463e-8ef9-140e21a515d4","sender":"User","text":"I want to create an agent for my company in LibreChat (github.com/danny-avila/librechat). This agent should be connected to an online repository. Users in my company should be able to add files to this repository, which in turn should be available for the agent to use as knowledge. Could we use MCP for this? Could we build a script that updates this repository to the agent files? What are my options? Do a deep research.","timestamp":"2025-08-07T10:01:37.304Z","tokenCount":96}
{"level":"debug","message":"[saveConvo] api/app/clients/BaseClient.js - saveMessageToDatabase #saveConvo","timestamp":"2025-08-07T10:01:37.334Z"}
{"level":"debug","message":"[AgentController] Request closed","timestamp":"2025-08-07T10:03:17.396Z"}
{"level":"debug","message":"[AgentController] Request aborted on close","timestamp":"2025-08-07T10:03:17.400Z"}
{"level":"error","message":"[api/server/controllers/agents/client.js #sendCompletion] Operation aborted Aborted","stack":"Error: Aborted\n    at EventTarget.listener (/app/node_modules/@langchain/core/dist/utils/signal.cjs:20:24)\n    at [nodejs.internal.kHybridDispatch] (node:internal/event_target:831:20)\n    at EventTarget.dispatchEvent (node:internal/event_target:766:26)\n    at abortSignal (node:internal/abort_controller:370:10)\n    at AbortController.abort (node:internal/abort_controller:392:5)\n    at EventTarget.listener (/app/node_modules/@langchain/langgraph/dist/pregel/utils/index.cjs:88:28)\n    at [nodejs.internal.kHybridDispatch] (node:internal/event_target:831:20)\n    at EventTarget.dispatchEvent (node:internal/event_target:766:26)\n    at abortSignal (node:internal/abort_controller:370:10)\n    at node:internal/abort_controller:373:20","timestamp":"2025-08-07T10:03:17.404Z"}

Screenshots

Code of Conduct

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

[danny-avila]

Your connection is getting closed to the backend server, this is normal for long (more than 1 minute-long) http requests, and usually reverse proxy setups (NGINX/Traefik/Cloudflare) will have default timeouts.

Similar discussion:
#4111

[MTR3K]

@danny-avila regardless of the server timeout that can be fixed - would it make sense to have a dedicated research mode in the ui that makes the process of waiting for these deep research apis more interactive?

Like polling the responses endpoints of (azure) open ai for example. They indicate things like reasoning, internal mcp calls, web search. Or even going so far as to come back to the specific conversation later and find the actual response once it arrives?

just thinking off the top of my head here as I am finding myself building a separate deep research fronted to interact with these apis:

https://cookbook.openai.com/examples/deep_research_api/introduction_to_deep_research_api

[jcbartle]

@danny-avila - the proper mechanism - if the LibreChat project does want to support deep research modes - is probably to implement support for background mode + polling:

https://developers.openai.com/api/docs/guides/background

This is what we're doing outside the LibreChat project.