• v0.12.0 c519ee522f

    v0.12.0 — Recording pipeline reliability
    Some checks failed
    Release / Create GitHub Release (push) Failing after 5s
    Stable

    rbrooks released this 2026-06-23 00:24:57 +00:00 | 311 commits to main since this release

    Hardening of the voice-recording pipeline plus a new local LLM backend.

    Recording / audio

    • Enable members intent so voice-recv can resolve audio speakers
    • Per-packet Opus decode in the sink to survive corrupted/DTX/silence frames; decoder reset on error to prevent cascade failures
    • Pre-strip WebRTC extension headers (0xBEDE and non-0xBEDE prefixes) before Opus decode
    • Preserve audio buffers after stop_listening() so processing can read them
    • Adopt discord-ext-voice-recv PR #56 (DAVE fix), pinned to an exact commit
    • WhisperX custom server endpoints for transcription

    LLM

    • llama.cpp support for summarisation and Q&A; raise on empty LLM summary and log finish_reason

    Infra

    • NullPool engine in Celery tasks to avoid asyncpg event-loop conflicts
    • Install git in the bot image for pip git+ dependencies

    Retroactive release tagged at c519ee5 (2026-03-25).

    Downloads