Repository navigation
v2 provider: HTTP 400 from Copilot model restarts active session server #42089
Description
Summary
An active session repeatedly restarts when GitHub Copilot rejects gpt-5.6-sol on the /chat/completions endpoint. This interrupts tool calls and requires the user to ask the agent to continue.
Environment
- opencode version:
0.0.0-next-17297 - OS: Darwin 25.6.0, arm64
- Terminal: Ghostty (
TERM=xterm-256color,COLORTERM=truecolor) - Shell:
/bin/zsh - Install/channel: v2 next/dev runtime
- Active plugins:
boxedcode-atlassian-mcp.tsboxedcode-box.tsboxedcode-config.tsboxedcode-local.tsboxedcode-task.tszzz.tsjj.tsboxedcode.js
Reproduction
- Configure an agent with
"model": "github-copilot/gpt-5.6-sol". - Start a session and perform a multi-step task involving tool calls.
- Continue interacting with the session until the provider rejects a request to
/chat/completions.
Expected Behavior
Either route the model through a supported endpoint or report the provider error without restarting the active session server. Tool and conversation state should remain usable.
Actual Behavior
The server restarts and interrupts the current operation. This happened repeatedly in the same session, including while invoking an interactive question tool.
The server log contains:
Failed to drain Session
AI.Error: RequestExecutor.execute: Provider request failed with HTTP 400:
model "gpt-5.6-sol" is not accessible via the /chat/completions endpoint
Relevant sanitized stack:
at RequestExecutor.execute (ai/src/route/executor.ts:244:25)
at SessionRunner.callModel (core/src/session/runner/llm.ts:191:32)
at SessionRunner.runSteps (core/src/session/runner/llm.ts:133:16)
at SessionRunner.drain (core/src/session/runner/llm.ts:124:26)
at ServerProcess.start (cli/src/server-process.ts:73:29)
Additional Context
- The configured provider is
github-copilot. - The configured agent variant is
medium. - The failure occurred repeatedly during one session rather than only at startup.
- After each restart, the user was shown: “The server restarted while you were working. Continue from where you left off without repeating completed work.”
- No workaround was attempted during the session.
Activity
github-actions commentedon Aug 12, 2026
This issue might be a duplicate of or related to existing issues. Please check:
Failed to drain Sessionlog line and silent failure symptom overlap closely. The key distinction in your report is that the drain failure also triggers a server restart, which may be a separate bug.santhiprakash commentedon Aug 13, 2026
Hi — I'd like to work on this one. I traced the failure path to the v2 session runner in packages/core/src/session/runner/llm.ts: after publisher.failAssistant durably records the provider AIError (e.g. the Copilot HTTP 400), the function still falls through to
return yield* Effect.failCause(stream.cause), which re-throws the raw provider error through SessionRunner.drain. I will open a PR that stops the runner from re-throwing an already-recorded provider failure and instead lets it surface as the durable step failure it already published.stevoland commentedon Aug 13, 2026
fyi: I think there's a couple of things here that I poorly reported.
Root cause of the service restart was my plugin throwing an exception in
setup.I'm no longer seeing the
/chat/completionserror on 0.0.0-next-17403stevoland commentedon Aug 19, 2026
gonna close cos root cause was my bad, although I think exception should be handled better, things moving fast