v0.1.1 — first public release

Operating memory and behaviour for AI agents. Works on any project, fresh or existing.

This is the first public release of kaora-memory on PyPI.

Install

pip install kaora-memory
cd myproject
kaora init

Open Claude Code (or Codex, Cursor, Gemini CLI) in the project and type go. The agent reads the operating memory before doing anything else.

What's in v0.1.1

• kaora init — installs the 14-file canonical template (AGENTS.md, CLAUDE.md, AGENT_BRIEF.md, docs/*, .claude/settings.json + 2 hooks, .gitignore) with brownfield-safe policy.
• kaora check — operating-memory integrity linter, 6 categories (structure, ADR-005, ADR state, placeholders, hooks, settings).
• Brownfield safety: .kaora-bak rotation, surfaced merge warnings, .gitignore shipped by default.

Brownfield safety hardening (F1+F2+F3)

Before publishing, a parallel code-review + security-audit pass on the package surfaced 3 issues against the "Works on any project, fresh or existing" promise. All three fixed in 0.1.1:

• F1 — .kaora-bak now rotates to .kaora-bak.1, .kaora-bak.2, ... on subsequent kaora init runs. The original pre-kaora backup is never overwritten.
• F2 — merge failures on existing .claude/settings.json now surface a warnings (N): block in the install report. Silent skip removed.
• F3 — template/.gitignore shipped with logs/, .env*, *.pem, *.pypirc, plus standard Python/IDE defaults. Greenfield projects are protected by default from accidentally committing bearer tokens captured by the API-call hook.

Suite: pytest 68/68 green · kaora check . 0 ERROR / 0 WARN · twine check PASSED.

Links

• PyPI: https://pypi.org/project/kaora-memory/0.1.1/
• Philosophy: https://github.com/alex-lamport/kaora-memory/blob/main/docs/PHILOSOPHY.md
• Dogfooding report: https://github.com/alex-lamport/kaora-memory/blob/main/docs/DOGFOODING_REPORT.md
• ADR log: https://github.com/alex-lamport/kaora-memory/blob/main/docs/DECISIONS.md
• License: MIT

kaora-memory doesn't add intelligence to the agent. It adds preparation.