hello! this message keeps popping up regardless of my prompts: No stream activity has been received for the last40seconds. The connection may have stalled. Stop and retry the response. If this keeps happening and your deployment uses a load balancer, verify that its streaming timeout is at least 5 minutes.Learn more in our documentation

I tried leaving a prompt running for 20 min and still no progress

Hey! That timeout message usually means the underlying model call is hanging before sending any response. A few things to check: 1. *Which model are you using? Some models (especially self-hosted or certain providers) can stall on large prompts 2. Is this happening on all prompts or just specific ones? Try a simple "say hi" to isolate the issue 3. Are you using any tools that might be timing out?* A tool that hangs will block the whole response If you're on our hosted platform, which agent/app is this happening with? I can help dig into logs. 🤖 Slack Agent

I'm using GPT 5.6-sol, I'm on the hosted platform through docker, archestra on local host won't even let me stop the prompt to test out "say hi"

That's odd — GPT 5.6-sol should respond quickly on hosted. A few clarifications: 1. *"Hosted platform through docker" — are you running Archestra locally via Docker, or using our cloud-hosted version at archestra.ai? 2. If local Docker*: the timeout is likely your setup. Check: - docker logs <container> for errors - Your OPENAI_API_KEY env var is set correctly - Network connectivity from the container 3. If you can't even click "Stop" — that sounds like a browser/UI freeze. Try hard-refreshing (Cmd+Shift+R or Ctrl+Shift+R). Which scenario matches yours? 🤖 Slack Agent

running archestra locally via docker, running it on my MacBook right now

Got it — local Docker on Mac. Check these: 1. *Logs first: `` docker logs archestra `` (or whatever your container name is) — look for errors 2. API key set? Make sure OPENAI_API_KEY is passed to the container 3. Model access:* GPT 5.6-sol requires API access — confirm your OpenAI account has it enabled What do the logs show? 🤖 Slack Agent

I’m running Archestra locally in Docker on macOS. The container is healthy, and I’m using the ChatGPT Subscription/Codex provider rather than an OPENAI_API_KEY.

The issue is not a provider-authentication error in the logs. The UI gets stuck on an active run/tool call, shows “No stream activity has been received for the last 40 seconds,” and sometimes the Stop button appears unresponsive in the UI.

I reproduced it and checked docker logs. The relevant behavior is:

• the app and conversation APIs return 200

• the active-run endpoint continues being polled

• several file/artifact requests take 7–11 seconds

• the stop endpoint does work server-side and returns 200:

POST /api/chat/conversations/0a7311cf-e4a0-4227-916d-f17d58e7b8b1/stop

statusCode: 200

So the backend accepts the stop request, but the frontend remains stuck or does not update correctly.

I’m using:

• Archestra container:archestra/platform:latest

• local URL: <http://localhost:3000>

• container status: healthy

• model: GPT-5.6 Sol through ChatGPT Subscription/Codex

• conversation ID: 0a7311cf-e4a0-4227-916d-f17d58e7b8b1

• app ID: 8f32d537-c4ef-4a74-b108-61d512e92c0e

There are no 401, 403, 429, or provider API-key errors in this log segment. Could this be a stuck tool execution or frontend active-run state issue? Is there a known fix or a way to clear the active run without losing the app/session?

Thanks for the detailed debug — this is helpful. You've nailed it: backend stops fine (200), but frontend stays stuck polling the active-run. This is a known frontend state sync issue when a tool execution hangs mid-stream. *Quick fix to unstick it: 1. Hard refresh the browser (Cmd+Shift+R) — this forces the UI to re-fetch conversation state 2. If that doesn't work, open a new browser tab to localhost:3000 and navigate back to the app To clear the active run via API (if UI stays stuck): ``bash curl -X DELETE <http://localhost:3000/api/chat/conversations/0a7311cf-e4a0-4227-916d-f17d58e7b8b1/active-run> `` Root cause: The 7–11 second file/artifact requests suggest a tool is doing heavy I/O and the stream times out before completion. The stop request kills the backend process, but the frontend's polling loop doesn't always catch the state change cleanly. For future runs:* If you're doing file-heavy operations, the ChatGPT Subscription/Codex provider can be slower than direct API — it proxies through their infrastructure. Consider using an OPENAI_API_KEY directly if you have one. Did the hard refresh or DELETE call unstick it? 🤖 Slack Agent