You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The description is too vague to be actionable — "The agent is stuck at thinking and not giving any response" does not provide enough context to reproduce or investigate the problem.
OpenCode version is missing (required).
Steps to reproduce are missing (required for bug reports).
Operating System and Terminal are missing.
Please edit this issue to fill in the missing fields within 2 hours, or it will be automatically closed.
This issue might be a duplicate of existing issues. Please check:
This has been occurring since 21:30. When I set deepseek-v4-flash as the default request model and keep only deepseek-v4-flash in the candidate models, it keeps showing 'Thinking...'. However, when I add glm5.2 to the candidate models, even though my default model and the top-priority candidate model are still deepseek-v4-flash, I see in the OpenCode Go backend that the model being requested is actually glm5.2.and I also tried for mimo-v2.5,it's same as deepseek-v4-flash
I use opencode in Home assistant and see the same behaviour. I have tried attaching more useful info:
[Bug] OpenChamber / OpenCode crash on restart after manual configuration reload ("OpenCode service unavailable" & WebSocket/SSE proxy errors)
Summary
When triggering a Manual configuration reload, OpenChamber attempts to restart the managed OpenCode instance. During this process, existing connection proxies (SSE and WebSockets) crash, throwing UND_ERR_SOCKET / terminated errors, followed by Error: OpenCode service unavailable while OpenChamber tries to establish connections before the new OpenCode process is fully ready.
Although OpenChamber eventually recovers once the port is detected and [PushWatcher] reconnects, the transition generates substantial error noise and stream disconnections.
Additionally, there is continuous log spam regarding Git operations (simple-git) failing repeatedly when running in non-git directories (such as /data, which is simply used to store session files).
Environment / Context
Environment: Home Assistant Add-on container (s6-overlay / s6-rc)
Interface Mode:openchamber
OpenCode Version:1.18.10 (also observed on 1.18.9)
Node module path:@openchamber/web
Working Directory:/data (non-git directory containing session files)
Steps to Reproduce
Start the add-on with INTERFACE_MODE=openchamber in a non-git working directory (e.g., /data).
Trigger a Manual configuration reload via settings.
OpenChamber attempts to stop the running OpenCode process and launch a new one on a newly allocated port.
Observe the log output during the transition.
Observed Behavior
Abrupt termination of SSE/WS streams:
When OpenCode is stopped during the config reload, OpenChamber encounters unhandled socket errors from undici:
[proxy] OpenCode SSE proxy error: terminated
Message stream WS proxy error: TypeError: terminated
at Fetch.onAborted (node:internal/deps/undici/undici:11700:53)
...
[cause]: SocketError: other side closed (UND_ERR_SOCKET)
Race condition during service startup:
While launching the new OpenCode managed server on the new port (e.g., 46209), OpenChamber repeatedly attempts to connect before OpenCode is fully bound and ready, leading to temporary connection error spikes:
Message stream WS proxy error: Error: OpenCode service unavailable
at buildUrl (file:///usr/local/lib/node_modules/@openchamber/web/server/lib/event-stream/global-hub.js:73:17)
at file:///usr/local/lib/node_modules/@openchamber/web/server/lib/event-stream/upstream-reader.js:124:23
[PushWatcher] disconnected OpenCode service unavailable
(Note: The system eventually recovers after port detection, but the process startup window is unhandled.)
Excessive Git Discovery / Worktree Error Spam:
When cwd is set to a non-git directory like /data, simple-git continuously attempts rev-parse --show-toplevel and worktree listings, flooding the log stream:
Failed to get remotes: GitError: fatal: not a git repository (or any parent up to mount point /)
Stopping at filesystem boundary (GIT_DISCOVERY_ACROSS_FILESYSTEM not set).
Expected Behavior
Graceful Proxy Teardown: SSE and WebSocket proxy connections should handle upstream process termination gracefully without throwing unhandled socket errors.
Readiness Polling: OpenChamber should wait for the new OpenCode HTTP/API server readiness check before attempting to reconnect PushWatcher / WebSocket streams.
Git Check Guard: OpenChamber / simple-git should verify if the working directory is a valid Git repository before executing Git commands to avoid continuous log spam in non-git directories.
Activity
github-actions commentedon Aug 4, 2026
This issue doesn't fully meet our contributing guidelines.
What needs to be fixed:
Please edit this issue to fill in the missing fields within 2 hours, or it will be automatically closed.
This issue might be a duplicate of existing issues. Please check:
If you believe this was flagged incorrectly, please let a maintainer know.
marvix1122333 commentedon Aug 4, 2026
same here, even ordinary "hello" , i didnt get any response
asim-ali-peerzada commentedon Aug 4, 2026
Same here resolve it ASAP!
Serpentv commentedon Aug 4, 2026
me too, it still thinking for 15 minute
justdoshiba commentedon Aug 4, 2026
same, only on deepseek v4 flash new free, other models seem to work correctly
PhucTruong-ctrl commentedon Aug 4, 2026
+1 with Deepseek v4 flash new, encountering this for like 1 hour and still not working.
DejanGegic commentedon Aug 4, 2026
can confirm, this has been going on for a bit longer than this issue is opened.
I tested multiple models on Zen and Go, it's a DeepSeek V4 Flash specific issue as i see. Kimi K3 has some "hickups" too but it's good enough
DejanGegic commentedon Aug 4, 2026
We are paying customers, but keep your tone polite and respectful, thank you.
lightninng commentedon Aug 4, 2026
This has been occurring since 21:30. When I set deepseek-v4-flash as the default request model and keep only deepseek-v4-flash in the candidate models, it keeps showing 'Thinking...'. However, when I add glm5.2 to the candidate models, even though my default model and the top-priority candidate model are still deepseek-v4-flash, I see in the OpenCode Go backend that the model being requested is actually glm5.2.and I also tried for mimo-v2.5,it's same as deepseek-v4-flash
MrBoggi commentedon Aug 4, 2026
I use opencode in Home assistant and see the same behaviour. I have tried attaching more useful info:
[Bug] OpenChamber / OpenCode crash on restart after manual configuration reload ("OpenCode service unavailable" & WebSocket/SSE proxy errors)
Summary
When triggering a Manual configuration reload, OpenChamber attempts to restart the managed OpenCode instance. During this process, existing connection proxies (SSE and WebSockets) crash, throwing
UND_ERR_SOCKET/terminatederrors, followed byError: OpenCode service unavailablewhile OpenChamber tries to establish connections before the new OpenCode process is fully ready.Although OpenChamber eventually recovers once the port is detected and
[PushWatcher]reconnects, the transition generates substantial error noise and stream disconnections.Additionally, there is continuous log spam regarding Git operations (
simple-git) failing repeatedly when running in non-git directories (such as/data, which is simply used to store session files).Environment / Context
openchamber1.18.10(also observed on1.18.9)@openchamber/web/data(non-git directory containing session files)Steps to Reproduce
INTERFACE_MODE=openchamberin a non-git working directory (e.g.,/data).Observed Behavior
Abrupt termination of SSE/WS streams:
When OpenCode is stopped during the config reload, OpenChamber encounters unhandled socket errors from
undici:Race condition during service startup:
While launching the new OpenCode managed server on the new port (e.g.,
46209), OpenChamber repeatedly attempts to connect before OpenCode is fully bound and ready, leading to temporary connection error spikes:(Note: The system eventually recovers after port detection, but the process startup window is unhandled.)
Excessive Git Discovery / Worktree Error Spam:
When
cwdis set to a non-git directory like/data,simple-gitcontinuously attemptsrev-parse --show-topleveland worktree listings, flooding the log stream:Expected Behavior
simple-gitshould verify if the working directory is a valid Git repository before executing Git commands to avoid continuous log spam in non-git directories.Log Excerpt
marvix1122333 commentedon Aug 4, 2026
i think it is fixed for big pickle model
DejanGegic commentedon Aug 4, 2026
Thanks OpenCode team. This is fixed now.
github-actions commentedon Aug 4, 2026
This issue has been automatically closed because it was not updated to meet our contributing guidelines within the 2-hour window.
Feel free to open a new issue that follows our issue templates.