Make your computer a home for your AI partner. Avibe is the local-first Agent OS: Vibey, its built-in AI partner, works out of the box, and Claude Code, Codex & OpenCode run unchanged beside it — reach them from your browser, phone, or any chat app.
Minimal CI Artifact Builder
Scope and Evidence
This is phase two of the approved order: stability, build/install preparation, then repeated work inside test calls. Phase one (#1905) is merged and its exact master source passed all 17 checks, 405 unit-file exits/metrics, 21 distribution contracts and 196 installer cases. The remaining test-call investigation is not claimed complete by this PR.
The artifact producer installed the whole application and its runtime dependency graph before running isolated PEP 517 builds. Inspection of the manifest helper, Hatch hook and exact-peer metadata hook found only standard-library operations and declared build-system requirements. The producer does not run application behavior; separate private installation jobs own that validation.
A fresh local builder containing only pip, build, packaging and pyproject-hooks produced real core and Memory wheels and source distributions. Neither the app nor its runtime dependencies were installed in that builder before or after. The original manifest preparation validated the published runtime manifest. Both distribution pairs used the exact 3.0.99rc1 peer contract. All 21 existing actual distribution tests passed against those artifacts, including isolated source installation. The local UI placeholder does not substitute for CI's production UI build, Docker regressions or Windows installation.
Smallest Complete Change
Install only the existing build frontend in the producer. Each isolated build continues installing its declared backend requirements. Preserve the original manifest preparation, core/Memory wheel and sdist builds, contract versions, production UI, artifact identities and same-run consumers. No build isolation is disabled. No runner, dependency declaration, timeout, test selection or application behavior changes.
Workflow contracts pin the minimal command sequence and execute that sequence with a controlled Python command to prove failure propagation at every boundary. All existing unit/UI and distribution/install aggregates still fail closed.
Verification and Delivery
Run workflow, shard and fixture contracts, source packaging contracts, lint and package requirement checks. Require all 21 actual distribution contracts in the installer CI with no skips; retain the real released-wheel upgrade and Windows smoke. The full expected check set remains 17. Complete current-head bot review, every exact-head lint run, whole-PR thread inventory and clean integration gate are mandatory before guarded squash. Verify exact-source master CI afterward.
Baseline #1905 PR run 34008949861 took 416 seconds; producer 103 seconds and installer 302 seconds. Merged master 34009487671 took 442 seconds, with a Python shard as the tail. Prior master 34007354215 took 361 seconds with producer 122 seconds. These varying samples are not a stable speed guarantee. Compare actual build-step and build-to-install timestamps after this PR; the successful local build is capability evidence, not a claimed CI speedup.
Initial findings-bearing review-head inventory is zero. No allocated cross-lane identifiers, production operations or global environment changes are involved.
Local workflow/shard/fixture/source contracts passed 70 cases. Seven artifact-only cases skip when wheel paths are absent; all 21 distribution cases separately passed with the minimal builder's real artifacts (136.41 seconds). Ruff 0.4.9, all 84 private test-environment package requirements and whitespace checks pass.