claude-fafm-sdk 0.3.0

Portable, cross-vendor AI memory in .fafm — offline-first SDK; connect a free namepoint for full intel.

claude-fafm-sdk

· · · · memory.faf.one

Portable, cross-vendor AI memory in .fafm. Give an AI agent memory that versions with your project and moves between models — instead of being locked to one vendor.

Offline-first: the local Soul works with no account. Connect a free namepoint for the full intel (semantic recall, smart-merge) at personal scale.

Two profiles, one .fafm format. This is the knowledge profile (typed, cross-linked agent memory). For the voice profile — the Voice Memory Layer — see grok-faf-voice.

Install

uv add claude-fafm-sdk          # in a project (recommended)
pip3 install claude-fafm-sdk    # also works

30 seconds

uvx claude-fafm-sdk init                 # 🧬 create a portable soul
claude-fafm-sdk etch "ships uv-first"    # write a memory
claude-fafm-sdk recall uv                # recall it  (filters: --tag --type --priority)
claude-fafm-sdk ls                       # list every fact, ranked
claude-fafm-sdk forget install           # delete a fact by id

Five commands — init, etch, recall, ls, forget. Run claude-fafm-sdk --help (or <cmd> --help) for the full surface.

Hand soul.fafm to grok-faf-voice and it reads it — same format, no fork.

Quickstart

from claude_fafm_sdk import Soul

soul = Soul("@me")
soul.etch("ships uv-first", id="install", type="reference", priority="high")
soul.etch("portable across vendors", id="why", type="project")
soul.save("me.fafm")            # → application/vnd.fafm+yaml

# later, anywhere:
soul = Soul.load("me.fafm")
soul.recall("uv")               # deterministic recall, ranked by priority + recency

That's the whole offline loop — no account, no server.

More in examples/ — portability + a real cross-vendor roundtrip (SDK writes .fafm, grok-faf-voice reads it back).

Go cross-vendor (a namepoint)

A namepoint is your soul's live address — mcpaas.live/<handle> — readable by Grok and any model. Zero-config: just push. With no setup, the first push auto-provisions an anonymous namepoint and saves it locally — no claim page, no key to copy.

uv add "claude-fafm-sdk[namepoint]"      # hosted ops use the family MCP client
claude-fafm-sdk namepoint push           # → live at mcpaas.live/anon… (auto-provisioned)
claude-fafm-sdk namepoint pull           # merge your hosted soul back (public read)
claude-fafm-sdk namepoint sync           # reconcile both ways

A-for-first-touch, B-for-keepers. The anonymous namepoint is session-like — lose this machine and it orphans (a reminder of the statelessness memory cures). Make it permanent + recoverable:

claude-fafm-sdk namepoint claim --email [e-mail]   # a named, recoverable namepoint
claude-fafm-sdk namepoint status                           # what you've got

The whole .fafm document is stored at the namepoint, so ids, types, and priorities survive the round-trip — structured memory, not prose. Reads are public (no key); writes use the key auto-provisioning hands you. The local Soul still works fully offline, no account.

Programmatic access mirrors the CLI:

from claude_fafm_sdk import Namepoint, Soul, provision_anonymous

soul = Soul.load("soul.fafm")
ident = provision_anonymous()                       # {namepoint, key}, zero-config
await Namepoint(ident.namepoint, api_key=ident.api_key).replace(soul.to_yaml())
body = await Namepoint(ident.namepoint).pull()      # reads are public — no key

Why

AI memory is vendor-locked. .fafm is the open, portable format — and this SDK is the open, offline-first way to use it. Souls written here interop with the fafm-engine and grok-faf-voice implementations: one format, never a fork.

Citation

If you use claude-fafm-sdk or the .fafm format in research or production, please cite the format paper:

Wolfe, J. (2026). Permanent Memory and Instant Recall: The .fafm Standard for Multi-Profile AI Agent Memory. Zenodo. https://doi.org/10.5281/zenodo.20348942

BibTeX

@article{wolfe2026fafm,
  title     = {Permanent Memory and Instant Recall: The .fafm Standard for Multi-Profile AI Agent Memory},
  author    = {Wolfe, James},
  year      = {2026},
  month     = {may},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.20348942},
  url       = {https://doi.org/10.5281/zenodo.20348942}
}

DOI: 10.5281/zenodo.20348942

License

MIT. The format is open (spec); the SDK is open; the at-scale intel + hosting is the paid tier.