Skip to content

v1.6.1

Choose a tag to compare

@github-actions github-actions released this 09 Oct 14:18
· 19 commits to main since this release
4e0801f

Memory Vault v1.6.1

Dashboard Chat works again with local models that have no reasoning toggle.

Fixed

  • Chat failed with a 400 on models without reasoning configuration. Chat
    calls LM Studio's native /api/v1/chat first and sends "reasoning": "off"
    with it. Models that expose no reasoning configuration — Gemma among them —
    reject that parameter outright, and the fallback to the OpenAI-compatible
    /v1/chat/completions only ran on 404, 405 and 501, so the 400 surfaced as
    "LLM error" and Chat was unusable with those models. Search, ingestion and
    MCP recall were unaffected.

    The fallback now also covers 400: any rejection of the native payload moves
    the request to the compatible endpoint, which sends no reasoning field at
    all, so this covers other native-only parameters a future model may refuse
    rather than just this one. Each fallback is logged.

    Reported by @cosmicwombat (#245), with the cause and both affected call
    sites already identified.

  • Streaming chat errors now name the cause. /api/chat/stream reported every
    failure as "LLM error. Check server logs.", so the status code and endpoint
    the response already carried never reached the user — diagnosing the bug
    above meant reading the LLM server's own logs. Streaming errors now include
    both.

Upgrading

Nothing to do beyond pulling the new image. No configuration, schema or API
changes.