LAMF — Ledgered Agent Memory Fabric

A governed, local-first memory authority for AI agents.

One governed memory for every agent.

LAMF gives compatible AI agents a shared, durable memory on hardware you control. It combines revisioned memory records, an append-only integrity spine, security profiles, secret exclusion, scoped retrieval, and MCP integrations without making any single model vendor the authority.

This repository is a clean deployment distribution. It intentionally contains no live memory database, operator token, vault, credentials, logs, or configuration from the system used to develop LAMF.

Status

LAMF 2.0.0 is a reference implementation intended for evaluation and development. It has automated validation and smoke tests, but it is not yet presented as production-hardened software. Review the selected security profile and deployment boundary before storing important information.

Requirements

• Python 3.10 or newer
• Windows, macOS, or Linux
• Node.js 24 or newer only when using the native OpenClaw integration

Quick start

Clone or download this repository, then run the installer from the repository root.

Windows

powershell -ExecutionPolicy Bypass -File .\installer\Install-LAMF.ps1

macOS or Linux

bash ./installer/install.sh

The interactive installer creates the private LAMF authority outside this Git checkout, initializes the selected security profile, optionally configures agent harnesses, and can start the local service. Non-interactive installation defaults to the controlled profile, no Obsidian projection, and no agent harness.

For repeatable deployment examples and verification commands, see INSTALL.md.

For a synthetic cross-agent walkthrough, see the reproducible demo. Architecture, threat boundaries, and troubleshooting are documented separately so evaluators can review limitations before using important data.

What is included

• installer/ — cross-platform installer and uninstall helpers
• runtime/ — local authority, CLI, MCP server, web workspace, and tests
• 02_SECURITY/profiles/ — fixed security profiles
• 03_CONTRACTS/golden-vectors.json — canonical-hashing validation fixture
• 04_STORAGE/SCHEMA.sql — authoritative SQLite schema
• 05_INTEGRATIONS/ — native OpenClaw integration
• Credit.md — living external-attribution and provenance ledger

Optional agent optimizations

LAMF's memory authority works without agent-behavior optimizations. The independently switchable optimization modules are distributed separately so evaluators can choose the smallest deployment surface:

• Download LAMF Optimizations

Install the pack into this checkout, rerun LAMF, and use lamf optimizations status to inspect its modules. Removing the pack returns LAMF to its unoptimized core behavior without affecting durable memory.

Security boundary

Keep the Git checkout separate from every live LAMF data directory. LAMF authority data, tokens, keys, spool files, databases, vaults, logs, and exports must never be committed. The included .gitignore is defense in depth, not a substitute for reviewing staged files before every push.

LAMF listens on localhost by default. Do not expose its service port to a network without an explicit security review and an appropriate authentication boundary.

License and attribution

LAMF is released under the MIT License. External influences and license-review status are recorded in Credit.md. The optional pack maintains module-level provenance in its own living ledger.

Community

Read CONTRIBUTING.md before opening a pull request, use private vulnerability reporting for security concerns, and follow the code of conduct. Public launch material remains a draft until maintainer approval.