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
I searched existing issues and found no report for AutoClaw model requests returning HTTP 406.
I read CONTRIBUTING.md.
Category
Model config / model requests
Agent framework
Not framework-specific — AutoClaw Desktop
Severity
Blocking for the built-in model route
Reproducibility
Always in this installation; repeated on multiple attempts
Problem description
On Windows, AutoClaw's built-in Z.ai Auto Mode (zai_auto) fails on ordinary chat prompts. Signing out and signing back in did not resolve it.
The app log reports LLM request failed. rawError=406 status code (no body). The request reaches POST /autoclaw-proxy/proxy/autoclaw/chat/completions, which returns HTTP 406 with an empty response body. This was observed repeatedly on 2026-09-23 and 2026-09-24.
A separate minimal authenticated chat-completion probe also returned 406 with an empty body. Variations tested included streaming/non-streaming and model: auto/model: zai_auto. An unauthenticated control returned 401, not 406. No access token or account identifier is included here.
Steps to reproduce
Install/run AutoClaw Desktop 1.17.8.124 on Windows and sign in.
Select the built-in Z.ai Auto Mode (zai_auto).
Send a short prompt such as Reply with one word: OK.
Observe the request fail with HTTP 406 and no response body. Signing out and back in does not change the result.
Expected behavior
The built-in model returns a completion, or the app shows a useful error that identifies the actionable reason for rejection (for example, a model, account, or quota restriction).
Actual behavior
The backend returns HTTP 406 with zero response bytes, and the app surfaces only the generic 406 status code (no body) error. The exact server-side reason is unknown from the response.
Version and environment
AutoClaw Desktop: 1.17.8.124
OS: Windows
Built-in model/provider: zai_auto / Z.ai Auto Mode
First observed: 2026-09-23; reproduced 2026-09-24
Diagnostics
AutoClaw request ID from the local trace: af7194d3-ee7b-4369-ba7b-0f5c5940db0c
Startup auth/profile refresh logs also show getaddrinfo ENOTFOUND autoglm-api.autoglm.ai, followed by cached profile data and has_agent_code=false. This may be related, but is not confirmed as the cause.
At a later check, DNS resolution and TCP 443 connectivity succeeded while the authenticated model request still returned 406.
Earlier requests through a separately configured OpenRouter route returned HTTP 200 in the same runtime logs.
Please investigate why authenticated AutoClaw model requests receive an empty 406, and return an actionable diagnostic if the rejection is expected for this account/model/route.
Activity
github-actions commentedon Sep 24, 2026
👋 感谢你的反馈,我们已经收到。
status: 待评估,你可以随时补充信息。needs: 更多信息标签并 @ 你。👋 Thanks — we've received your issue.
status: 待评估(Triage); feel free to add context.needs: 更多信息and ping you.