engrim vs pro-workflow
Local SQLite episodic memory shared across Antigravity, Claude Code, Cursor, Windsurf and Codex: hooks/MCP capture decisions, FTS5+model2vec hybrid recall reloads ~1k tokens per session. — versus — One SQLite store under every Claude Code session: corrections become FTS5-searchable rules that auto-load, research grows persistent wikis, and 37 hook scripts add quality gates.
Both put an FTS5 SQLite store under coding-agent sessions and auto-load hits; pro-workflow is Claude-Code-only and correction-centric, engrim is harness-neutral and decision-centric.
| engrim | pro-workflow | |
|---|---|---|
| Stars | 222 | 2.8k |
| Forks | 12 | 285 |
| Language | Python | JavaScript |
| License | MIT | — |
| Last activity | yesterday | 11 days ago |
| Topics | memory, coding | coding, memory |
| Curated connections | 5 | 7 |
engrim — the curator's take
Use it when you actually switch harnesses mid-project — Gemini in Antigravity in the morning, Claude Code in the afternoon — and want one memory both write to and read from, with `origin_agent` provenance on every entry. `engrim setup` auto-wires hooks, MCP and status lines for whatever it finds under ~/. The retrieval is cheap and offline (FTS5 + static model2vec embeddings, no API), and the deliberate 4k-char 'memory pack' cap keeps it from becoming a second context window. Don't pick it for cross-project or team memory (it is per-project, per-machine), or if you want raw transcript capture rather than curated decisions/facts/feedback. Small project (~200 stars, one maintainer) — expect rough edges.
pro-workflow — the curator's take
The most complete attack on Claude Code amnesia we've mapped: correct it once and the correction becomes a durable, searchable rule; research lands in wikis that persist and even grow via an auto-research loop; hooks add git/secret guards and cost tracking. After 50 sessions the compounding is real. Two cautions: NO license file at review time (all rights reserved by default — same gap as this author's other tools), and 34 skills + 37 hooks is a lot of surface — adopt the memory core first, audit the hooks before letting them gate your commits.