Skip to content

v0.1.0

Choose a tag to compare

View all tags
@oscardvs oscardvs released this 03 Jul 16:20
v0.1.0
df0c962

Website: https://oscardvs.github.io/anamnesis/ · Docs: https://oscardvs.github.io/anamnesis/docs

uv tool install anamnesis-memory && anamnesis init

First launch release of the local-first core.

Highlights since 0.0.2

Memory quality loop

  • LLM reflection distills session logs into durable notes (any OpenAI-compatible provider, config-driven, swappable)
  • Recall-gated merge consolidates duplicates; nothing applies unless recall on your eval set holds
  • Provenance-aware recall: every generated note carries source + confidence; superseded notes drop out of injection
  • Durable writes: every path commits before it reindexes, so concurrent syncs cannot wipe work
  • Opt-in auto-reflect at SessionEnd once a project crosses the unreflected threshold

Foundation

  • Multi-tenant-shaped schema: user_id / workspace_id end to end (store, MCP tools, CLI)
  • Eval harness: recall@k + inject-size measurement, sandboxed before/after experiments (anamnesis eval)
  • Native importer mirrors Claude Code's own per-project memory into the store (anamnesis import)
  • anamnesis config CLI + persistent per-store config.json; dashboard settings page

Install hardening (from a clean-room stranger audit)

  • fastmcp is now a base dependency: the 0.0.2 one-line install registered an MCP server that could not start (0.0.1 and 0.0.2 are yanked)
  • A local-only store can attach a remote later and just sync
  • First sync of a fresh store is graceful and reports the history it pulled
  • Sync failures print one line instead of a traceback

Honest numbers

  • The token benchmark (bench/cross-machine-tokens/) is reproducible on a Pro/Max subscription, no API key needed. Measured: ~8% fewer input tokens on the same fresh-machine task, conventions known from the first turn. The earlier null result is published alongside.
Assets 2
Loading