RMS Memory Core

rms-memory-core is a public alpha, local-first, read-only memory projection shell.

Version: v0.1.0-alpha
Status: public alpha
Production ready: no

What it does

• exposes small, inspectable memory projections;
• evaluates local JSONL fixtures;
• records warnings and candidate feedback without promoting them to truth;
• provides a read-only MCP-shaped resource surface;
• keeps work evidence separate from companion/persona projections.

What it does not do

• it is not a production memory service;
• it does not dispatch, apply, commit, or mutate a canonical project truth;
• it does not promote a memory candidate automatically;
• it does not provide write-enabled MCP memory;
• it does not claim production readiness.

Quick start

Requires Python 3.11 or newer.

python -m pip install -e ".[test]"
python scripts/smoke_test.py
python -m rms_memory.cli hello
python -m rms_memory.cli inspect --domain work
python -m rms_memory.cli demo work-handoff
python -m pytest -q tests

The commands are local and deterministic. No network service is required by the package itself.

Design boundary

Projections, candidates, warnings, feedback, and retrieval rankings are evidence-bearing artifacts. They are not authority by themselves. See ARCHITECTURE.md, SECURITY.md, and KNOWN_LIMITATIONS.md.

License

Apache-2.0. See LICENSE and LICENSE_DECISION.md.