Skip to content

Releases: cremenescu/mem0ry4ai

v0.3.0 — automatic end-of-session git checkpoint

10 Jun 09:54
@cremenescu cremenescu
v0.3.0
83ed486

Choose a tag to compare

View all tags
v0.3.0 — automatic end-of-session git checkpoint Latest
Latest

New

  • Automatic git checkpoint at session end — the SessionEnd hook now commits store/ by itself (authored mem0ry4ai hook, no signing, silent on errors/locks). Memories written during a session land in git history without any manual action; the uncommitted changes state resolves itself when the session closes.
  • Health panel: an uncommitted store is now shown as an informative gray state ("auto-checkpoint at session end"), not a red error — it is the normal mid-session condition.
  • The commit button on the git history page remains for mid-session checkpoints with a custom message.

Full diff: v0.2.0...v0.3.0

Assets 2
Loading

v0.2.0 — git history page

10 Jun 09:06
@cremenescu cremenescu
v0.2.0
35ce836

Choose a tag to compare

View all tags
v0.2.0 — git history page

New

  • Git history page — the store's timeline in the web UI: every commit touching store/ listed with a lazy-loaded, colored diff. See exactly what was learned, superseded or removed, and when.
  • Commit from the UI — when the store has uncommitted changes, a card lists them and a button commits store files only (authored mem0ry4ai web, no signing, never touches code). Works when the server runs as your user (server_web.sh).
  • Bilingual like the rest of the UI (EN default / RO).

Docs

  • README: Measured impact section with real before/after numbers from the author's setup (88–92% less fixed context per session) and honest caveats.
  • README gallery: web UI screenshots with demo data, including the new git history page.

Full diff: v0.1.0...v0.2.0

Loading

v0.1.0 — first public release

10 Jun 07:44
@cremenescu cremenescu
v0.1.0
28800ad

Choose a tag to compare

View all tags
v0.1.0 — first public release

First public release of mem0ry4ai — persistent, local-first memory for coding agents (built for Claude Code).

Highlights

  • Markdown + git as the source of truth — auditable, diffable, supersede-never-delete; the SQLite FTS5 index is derived and disposable.
  • Deterministic Claude Code hooks: inject relevant memories at SessionStart (scoped per project; capped multi-project index from a monorepo root), capture transcript pointers at SessionEnd/PreCompact.
  • Trust-gated capture: the in-context agent writes directly; optional offline extraction with a local LLM (Ollama) goes through a human review queue — small models are noisy and over-confident (we measured ~1.0 confidence on everything).
  • First-class todo and status types — they answer "where was I?" when you return to a project after weeks, pinned first in injection and UI.
  • Web UI (standalone php -S launcher, auto-started by the session hook): system dashboard with health checks and live updates, per-project pages, ranked FTS5 search, bulk operations, supersede-chain navigation, a "What Claude sees" injection preview, and the review queue. Bilingual (EN default / RO).
  • Zero heavy dependencies: Python stdlib + PHP 8 + git. No Docker, no vector database, no API keys.

Install

git clone https://github.com/cremenescu/mem0ry4ai.git && cd mem0ry4ai
./mem.py list                      # CLI works immediately
./server_web.sh                    # web UI -> http://127.0.0.1:8841/
python3 hooks/install.py --target user   # Claude Code integration

See the README for the full story, including why this exists and what we learned surveying the existing memory-tool landscape.

Loading