Stuck on “Brainstorming” with LM Studio + Qwen3.6‑27B (skipSearch=true, empty content)

Description Vane 1.12.2 is connected to LM Studio via Docker on macOS. LM Studio receives requests and returns completions, but the UI in the “Everyday Research” profile gets stuck on the first “Brainstorming” phase and never shows an answer, even for very simple prompts like “Say hello only.”

Environment

Vane version: 1.12.2 (Settings → Version)

Install method: git clone of https://github.com/ItzCrazyKns/Vane.git + docker compose up -d

OS: macOS (Apple Silicon MacBook Pro)

LM Studio:

Local server port: 1234

“Serve on Local Network”: ON (server reachable at http://10.0.0.233:1234)

Model: Qwen/Qwen3.6‑27B (MLX, 4‑bit)

Vane LM Studio connection:

Base URL: http://host.docker.internal:1234/v1

Profile: “LM Studio – Everyday Research”

What I did

Started LM Studio, loaded Qwen3.6‑27B, and confirmed the local server is running on port 1234.

Started Vane from the cloned repo with docker compose up -d.

In Vane Settings → Models, created/edited an LM Studio connection with base URL http://host.docker.internal:1234/v1.

Selected that connection in the “Everyday Research” profile.

Asked a simple query: Say hello only.

What I expected Vane should show a short answer (“hello”) and optionally skip search, since the model doesn’t need external data for this question.

What actually happens The UI stays on “Brainstorming” and never moves on to search or answer.

However, LM Studio’s Developer Logs show that the request completed successfully.

LM Studio log snippet Here is an example of what LM Studio returns for the “Say hello” prompt:

{ "id": "chatcmpl-bd5jgsksbf4zzb6fogggs", "object": "chat.completion", "model": "qwen/qwen3.6-27b", "choices": [ { "index": 0, "message": { "role": "assistant", "content": "", "reasoning_content": "{"classification": {"skipSearch": true, "personalSearch": false, "academicSearch": false, "discussionSearch": false, "showWeatherWidget": false, "showStockWidget": false, "showCalculationWidget": false}, "standaloneFollowUp": "Say hello"}", "tool_calls": [] }, "finish_reason": "stop" } ], "usage": { "prompt_tokens": 1356, "completion_tokens": 60, "total_tokens": 1416, "completion_tokens_details": { "reasoning_tokens": 59 } }, "system_fingerprint": "qwen/qwen3.6-27b" }

So LM Studio is:

Receiving the request.

Generating structured reasoning_content with "skipSearch": true.

Returning an empty content string.

From the user’s perspective, Vane never leaves “Brainstorming,” even though the model has already decided to skip search and provided a follow‑up.

Additional notes I can see multiple LM Studio log entries for Vane requests, so the network wiring (Docker + host.docker.internal) appears correct.

The version is 1.12.2, so this is happening on a current build.

I’m happy to test any custom config or debug builds if that helps.

Thanks Pixelpossum