TL;DR: .faf is context. .fafm is memory. v7.2.0 puts portable agent memory in the CLI you already run — convert a Claude Code memory directory to a soul.fafm, then etch, list, and recall with the same INTEROP as claude-fafm-sdk 1.0.

npm install -g faf-cli@7.2.0

.faf is context. .fafm is memory.

For a year, faf-cli authored the context layer: one scored project.faf → AGENTS.md · CLAUDE.md · the whole vendor surface. That's the foundation AI needs to understand your repo.

Memory is the other half. Agents that forget between sessions re-ask the same questions and re-learn the same facts. The IANA-registered .fafm format (application/vnd.fafm+yaml) is how those facts travel — offline-first YAML, knowledge profile, cross-vendor. Python shipped the bar in claude-fafm-sdk 1.0. Today the CLI carries the same shape.

What's New

faf memory convert — Claude Code memory directory → soul.fafm. Schema-constrained (type gate, whitelist extraction — no frontmatter leak). Provenance rides in fact extras, not a forked schema.
etch · ls · recall · show — write a fact, list ranked, deterministic filter + rank, soul summary. --json when you need a pipe.
TS library — Soul + fromClaudeDir in the package root exports. Same INTEROP as the Python SDK: residual preserve, index rebuild, recall source of truth.
.fafm in faf formats — discoverable next to .faf and .fafb.

Try It

npm install -g faf-cli@7.2.0 faf memory convert ~/.claude/projects/.../memory -o soul.fafm faf memory recall "your query"

Prefer Python? Same INTEROP, same soul shape:

pip install claude-fafm-sdk==1.0.0 # or: uvx claude-fafm-sdk

Why both

Context is what the project is. Memory is what the agent learned. One without the other leaves the AI guessing or re-explaining. Two formats, one ecosystem, both IANA-registered — and now both reachable from faf.

The Numbers

v7.2.0 — released July 22, 2026
1,164 / 1,164 — tests passing
100% — Trophy FAF score
INTEROP — aligned with claude-fafm-sdk 1.0
IANA — .faf + .fafm both registered
Ecosystem downloads — live hard floor at faf.one/downloads

If you like our work, consider a star on the repo — it helps others find us too.

Star faf-cli