FAF Memory — Permanent Memory Layer for Claude Code

Your Claude doesn't remember yesterday. .fafm is the file that fixes that — IANA-registered, cross-vendor, offline-first. Open the same memory in grok-faf-voice — same facts, both directions tested.

Install

Not in Anthropic's directory yet (review pending). Until then, install from the FAF marketplace:

claude plugin marketplace add Wolfe-Jam/faf-plugins
claude plugin install faf-memory@faf-plugins

The plugin wires the faf-memory-mcp server into Claude Code via uvx, exposing five tools.

What it runs

One local MCP server, faf-memory-mcp==0.1.1, started with uvx (downloaded from PyPI on first run). It reads and writes a .fafm file on your machine and makes no network calls of its own. The plugin collects no data. Support: team@faf.one.

Prerequisite: uv — the plugin launches its MCP server with uvx. First run builds dependencies (~90s); warm starts are instant.

What you get

Tool	What it does
etch(text, id?, type?, priority?, tags?)	Write a durable fact. O(1) dedup by id.
recall(query?, type?, tags?, min_priority?, limit?)	Filter (substring + type + tags + priority floor), rank by priority then recency.
list_facts()	Enumerate all facts (no filter, no rank).
save_soul(path?)	Persist to a .fafm file.
load_soul(path?)	Load a .fafm from disk.

Memory lives in .fafm — plain YAML, IANA-registered. Diffable like code. Portable to any tool that reads the format.

How to know it's working

• Your agent doesn't ask "what is this project?" twice in the same week.
• recall("X") surfaces what you etched last session, not just the current one.
• The .fafm file grows readably — open it, diff it, share it.
• Open the same file in grok-faf-voice — same facts. Cross-vendor proven.

Complementary, not competing

.fafm is a structured source. It works alongside Claude Code's own memory, not against it:

• claude-md-management maintains the rendered CLAUDE.md; .fafm is its memory sibling
• Community remember keeps conversational daily logs; .fafm is its structured peer

PML is the standard underneath — IANA-registered, cross-vendor, offline-first.

The receipt

400+× faster type-filter queries vs grep on a real 492-file AI memory corpus. Falsifiable methodology + scripts + sanitized pilot at Wolfe-Jam/faf-memory-proof — reproduce in 30 seconds with one paste.

Tradeoff note

Biases toward deterministic recall (substring + type + tags + priority + recency) over semantic recall. For semantic / LLM smart-merge, see hosted namepoints in claude-fafm-sdk. Offline-first ≠ offline-only.

The FAF cluster

• This plugin — faf-memory — wraps faf-memory-mcp
• Wolfe-Jam/faf-memory-mcp — the MCP server
• Wolfe-Jam/faf-memory-proof — the falsifiable receipt
• Wolfe-Jam/faf-plugin — sibling: .faf context (FCL)
• claude-fafm-sdk — the open Python SDK
• Wolfe-Jam/faf — the format spec + IANA registration

License

MIT.