chore(release): v4.2.0 — AI Provider Abstraction & BYO-AI #504
Loading…
Reference in a new issue
No description provided.
Delete branch "chore/release-v4.2.0"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Release commit for the v4.2.0 milestone:
APP_VERSIONandBOT_EXPECTED_APP_VERSION→ 4.2.0 (scripts/check_version_sync.pypasses), changelog stamped with the preamble self-hosters read before upgrading.What ships (all merged since v4.1.2): provider contracts with swappable adapters — the bundled WhisperX server's API v2 (Rhoving/iac-repo#396) and an OpenAI-compatible ASR adapter covering OpenAI and Groq (#350, #496); per-campaign BYO-AI behind an instance switch (#128, #501); campaign vocabulary bias (#355, #491); real word timestamps carried through the pipeline (#352, #499); provider-aware concurrency caps with a queue the GM can see (#356, #495); per-session usage and cost with admin rollups (#357, #494, #503); capability gaps explained in plain language (#358, #492); a hardware preflight, a CPU profile and measured hardware profiles (#353, #354, #500); a conformance suite every adapter must pass plus a monthly canary (#359, #493); the committed synthetic session that tests transcription end to end (#433, #497); and the provider evaluation with its isolation test (#360, #498).
Five additive migrations, verified up/down/up:
a5b6c7d8e9fa→b6c7d8e9fa0b→c7d8e9fa0b1c→d8e9fa0b1c2d→e9fa0b1c2d3e, single head. Bot contract unchanged (v1); nothing under/api/bot/*changed shape.Operator notes carried in the preamble: update the bundled transcription server to API v2 for word timings and vocabulary bias (an older server keeps working, and the capability panel says to update); six new optional
.envsettings with safe defaults; ~9 MB of invented audio now in the repository.Deploy plan after tagging: dev first (with the #433 end-to-end live run through the real intake endpoint as part of validation), then production.
🤖 Generated with Claude Code
Deployed.
docker-test)docker-host-mcp)f7aa15e(tag), pulled toa483c87at 07:31 (evals-only, no rebuild)a483c87e9fa0b1c2d3e (head), 5 migrations fromf4a5b6c7d8eaVersion handshake OK: backend 4.2.0, bot API contract v1, gateway connectedDev was validated end to end before prod went out: two synthetic sessions (#433,
fullandic_only) submitted throughPOST /api/bot/sessions/{id}/audioand processed by the real pipeline against the bundled WhisperX v2 server and the llama.cpp endpoint. Capability discovery, the provider queue lease, usage rows, BYO-AI precedence and the monthly canary were all checked live. Numbers on #433. Three rough edges found, none affecting session processing, filed as #507 (preflight LLM probe on reasoning models), #508 (session_usage.asr_modelempty) and #506 item 5 (fixture links unverified).Operator note carried over from the release preamble: the bundled WhisperX server must be on API v2 (it is, both environments share the one at 10.3.0.28:8091).