v0.12.12
Vexa-ai/vexav0.12.12Jul 18, 2026by DmitriyG228
Upgrading?
AI Summary
A production-fix release addressing recording freezes, truncation issues, and Helm redundancy problems.
Key Highlights
- Recordings can freeze; masters are now re-assemblable.
- >1000-chunk recordings are now paginated to prevent silent truncation.
- `replicaCount>1` now provides real redundancy via topologySpreadConstraints.
- Helm upgrade deadlock on agent-api fixed.
- arm64 is now smoke-executed in every release run.
New Features
- Recording self-repair on read.
- Recording pagination.
- Helm topology constraints.
- arm64 smoke execution.
- Model ID lineage tracking.
Full Release Notes
## v0.12.12 — the production-fix release 14 PRs of the v0.12.11 candidate batch (absorbed per the two-phase flow — that candidate was tagged, validated, and superseded before witness) **plus five production fixes reported and shipped same-day**. ### Fixed, in production's own words - **Recordings could freeze at first read (#768, P0)** — opening a recording mid-meeting permanently truncated it (observed in prod: a 4-hour meeting serving 6.6% of its audio). Masters are now re-assemblable; **already-frozen masters self-repair on the next read** — no data was lost, chunks were always stored. - **>1000-chunk recordings silently truncated (#769)** — chunk listing paginates to exhaustion, with loud truncation guards. - **`replicaCount>1` bought no real redundancy (#770)** — the chart now expresses `topologySpreadConstraints` (global + per-component, each component's own selector injected). - **`helm upgrade` deadlocked agent-api (#774)** — RollingUpdate surge vs its RWO workspace PVC; now `Recreate` on RWO, RollingUpdate restored automatically under RWX. - **`BOT_SPEAKER_*` tuning was dead config (#771)** — the runtime kernel never merged the profile's base env into spawns; chart-set tuning now reaches every bot. ### Witnessed live on the delivered deployment Signed receipt: `releases/v0.12.12/witness.json` — 22 values, 4 walked live (including the #768 P0: mid-meeting master 6,959,152 B → post-completion 8,117,085 B — re-assembly proven on a live meeting), 18 by-proxy with named evidence. ### Also in this release - feat(release): native arm64 smoke in release-validate (#630) - fix(db-budget-gate): the pool scan counts production source — make the test filter express it (stacked on #703) (#704) - test(admin-api): inline the pool literals — the budget gate excludes tests by contract (#702) (#705) - governance(D-L4): the witness walks a DELIVERED deployment — enforced at the receipt (#709) - release: v0.12.10 witness receipt — SIGNED (human witnessed the assembled value) (#711) - governance: merge-card acceptance row — Closes cannot silently drop delivery legs (#719) - fix(release-witness-gate): full 64-hex image digests or no pass — truncation rule + v0.12.10 record-completion (#720) - docs(releases): step 4 — the publish act joins the runbook (gh release create + the guard it triggers) (#721) - harness: least-privilege env for the model subprocess (scrub REDIS_URL + dispatch token) (#726) - feat(gmeet): every segment carries the STT-detected window language (#523) (#760) - fix(meetings): pre-active teardown attributes the reason from the status (#598) (#762) - ci(pr-value): lite-smoke leg — build the PR lite image, boot it, smoke the front doors (#581) (#763) - feat(deploy): make probe — standing full-journey smoke probe per surface (#690) (#764) - fix(api.v1): 61 secured ops referenced a nonexistent APIKeyHeader scheme — resolve to ApiKeyAuth/AdminApiKeyAuth (#531) (#765) - release: fold #765's fragment — v0.12.11 batch completion (#772) - fix(runtime): kernel merges Profile.base_env into the spawn env — chart-set BOT_SPEAKER_* tuning reaches the bot (#771) (#775) - feat(helm): topologySpreadConstraints (global + per-component, own-selector injected) · agent-api Recreate on RWO workspace PVC (#770 #774) (#777) - fix(lite-smoke): bounded wait for internal admin-api — the leg's first real firing raced the boot (#780) ### Operator notes - **Rolling-upgrade note for `TRANSCRIPTION_MODEL`**: set the new env only after bot images are on v0.12.12 — older bots reject invocations carrying unknown fields (sealed-schema validation). - arm64 is now smoke-executed on native arm runners in every release run (#630 / #564). ### Credits Reporters & authors this release stands on: the production incident review (#768 #769 #770 #771 #774 #779 #783), @Som23Git (#630 arm64 smoke), @waqaskhan137 (#489 → #522 model-id lineage), @mnofresno (#607/#621 seam work), @rennf93 (#573 lineage), and the standing reporter cohort. The witness walk and value sign-offs are recorded in the receipt.
Related Repositories
public-apis/public-apis★ 491,256
A collective list of free APIs
NousResearch/hermes-agent★ 256,127
The agent that grows with you
yt-dlp/yt-dlp★ 200,304
A feature-rich command-line audio/video downloader
microsoft/markitdown★ 191,593
Python tool for converting files and office documents to Markdown.
Significant-Gravitas/AutoGPT★ 187,703
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.