Skip to content

Releases

v0.2.0

Choose a tag to compare

@radys-s radys-s released this 19 Nov 03:20
45088e2

Developer-experience release: AI Bridge is easier to start with and safer when several agents use it at once.

Fixes

  • Candidate review is atomic. Approving the same candidate from several agents at once used to create duplicate memories; now exactly one approval wins and the others get candidate_not_pending (HTTP 409). An approved candidate can no longer be rejected afterwards, and global-scope candidates can be approved.
  • Export includes every memory. aibridge export stopped at 100 memories without saying so.
  • No more tracebacks or 500s for an unknown candidate or an unregistered export path.

Clearer errors

  • Every API error now has the same shape as the CLI's error[code]: {"code": "...", "detail": "..."}, with 404 for unknown projects, memories, candidates and runs.
  • Validation errors name the field and the limit, e.g. title: String should have at most 300 characters. Rejected values are never echoed back.
  • The API reports its real version instead of a fixed 0.1.0.

Docs

  • The README quick start now has a macOS/Linux version, and its example actually retrieves the stored memory.

Upgrading: no database migration. API clients that checked for 400 on an unregistered project_path now get 404, and over-long memory fields now get 422.

v0.1.0

Choose a tag to compare

@radys-s radys-s released this 14 Nov 19:26
c507ba3

First tagged release of AI Bridge: a local memory and context layer for AI coding agents that retrieves only what the current task needs.

What's in it

  • Memory kernel on SQLite (WAL, FTS5) with validated events and atomic projections.
  • Checksummed migrations, checked on every start.
  • Local API foundation.
  • Fix: opening the database from several processes at once no longer fails with "database is locked" on Windows.
  • CI: ruff, strict mypy and pytest on Ubuntu and Windows.