0.2.0 establishes the initial shape of ax: a local graph for understanding how AI coding sessions actually unfold.
The release focuses on ingesting the raw material that matters most: Claude Code transcripts, Codex sessions, installed skills, and git history. The CLI can then query that graph to answer practical questions about sessions, skill usage, repo-scoped recall, and self-improvement evidence.
Highlights
- Local transcript ingest for Claude Code and Codex.
- Skill and invocation tracking.
- Git history ingest for repository context.
- CLI surfaces for sessions, recall, skills, and grounded improvement tasks.
Why it matters
Autocomplete had a tight feedback loop. Agentic coding needs one too. This release is the base layer: durable local memory that can be queried, inspected, and improved without shipping private transcripts to a hosted service.