MoFlo – give Claude Code a memory
MoFlo makes Claude Code remember what it learns, check what it already knows before re-reading your files, and get smarter with every session — automatically. No configuration, no API keys, no cloud.
# install & initialize
npm install --save-dev moflo
flo init
Restart Claude Code — memory, indexing, gates, and routing are all active.
It remembers
Semantic memory with 384-dim on-device embeddings. Claude searches what it already knows before burning tokens exploring files — and durable lessons persist across sessions.
It gets smarter
Task outcomes feed back into a learned routing system, so Claude picks the right approach next time. SONA, ReasoningBank, and trajectory persistence, wired out of the box.
It stays yours
100% local. No API keys, no external services, no data leaving your machine. Built on Node’s native SQLite + HNSW vector search — nothing to compile.
What’s inside
Everything below is configured and functional the moment flo init finishes.
Why local-first?
Most AI tooling ships your context to a cloud service. MoFlo doesn’t. The memory database uses Node’s built-in SQLite engine — no better-sqlite3 bindings to compile, no WASM round-trip. Embeddings run on-device with all-MiniLM-L6-v2. Your code, your docs, and everything Claude learns stay on your machine.
Frequently asked
Does MoFlo send my code anywhere?
No. It’s fully local — no API keys, no cloud, no telemetry required.
Do I need a separate API key?
No. The embeddings and memory run on-device.
Which platforms does it support?
macOS, Linux, and Windows, identically.
How is this different from Claude Code’s built-in context?
Claude Code’s context resets each session; MoFlo gives it durable, searchable memory that persists and compounds across sessions.
Is it open source?
Yes, MIT-licensed. View the source on GitHub →
Give Claude Code a memory
One install, one flo init, and Claude stops forgetting.
MoFlo began as a fork of Ruflo / Claude Flow and has since diverged into an independent project. Built by Eric Swann.
