OCMF — OpenClaw Memory Fabric

OCMF is a cross-host memory layer for AI developer tools.

It gives Claude and Codex a shared, traceable memory for project decisions, constraints, preferences, and execution context — so you do not have to repeatedly restate the same information across tools.

Why OCMF exists

When you work across multiple AI tools, memory becomes fragmented:

• Claude remembers one set of decisions
• Codex sees a different local context
• important constraints get repeated, drift, or disappear
• tool switching breaks continuity

OCMF solves that by providing a unified memory fabric with:

• shared memory across supported hosts
• provenance-aware recall
• event-backed memory records
• a unified CLI for install / remember / recall
• auditable outputs and evidence-driven validation

What OCMF does

Shared memory across hosts

Claude and Codex can read from the same memory layer.

Provenance-aware recall

Recall results show where a memory came from, such as:

• From Claude
• From Codex

Unified CLI

A single ocmaf command handles:

• host install
• remember
• recall
• status
• doctor

Event-backed model

Memories are stored as traceable events instead of opaque blobs.

Cross-host context

OCMF is designed for workflows where more than one AI host participates in the same project lifecycle.

Current status

Area	Status
Trusted user journey	Verified in current environment
Development install	Available
Global CLI (ocmaf)	Available
Quickstart	Verified
Wheel install rehearsal	Verified
Formal release readiness	Completed
Formal PyPI release	Not published
TestPyPI rehearsal	In progress / environment-dependent
OpenClaw host	Blocked / TBD

Current boundary:

• CURRENT_ENV_STABLE = YES
• ROOT_CAUSE_IDENTIFIED = NO

This means the main user journey is working in the current environment, but unresolved lower-level root-cause analysis is still explicitly not claimed as complete.

Supported hosts

Host	Status	Production path	Auto-memory
Claude	Supported	A1 + B	Supported
Codex	Supported	C	Manual / explicit path
OpenClaw	Blocked / TBD	N/A	N/A

Host notes

• Claude currently has the strongest end-user path.
• Codex is supported through the explicit/manual MCP-oriented path.
• OpenClaw is not yet available as a supported host.

Who this is for

OCMF is designed for developers who use more than one AI tool in the same workflow and want:

• less context repetition
• shared project memory across hosts
• traceable provenance
• auditable CLI behavior
• a memory layer that can be validated instead of guessed

Installation

Prerequisites

• Python 3.11+
• pip
• macOS / Linux shell environment recommended for the current validated path

Development install

git clone git@github.com:cait52099/openclaw_memory_fabric.git
cd openclaw_memory_fabric
python3 -m pip install -e .