the body that grows
npm install -g meetsoma && soma
First run, it reads your workspace and writes its own identity.
Second run, it remembers.
Session 1 · creates .soma/, scans your repo, writes identity.md
you didn't configure anything
Session 2 · loads yesterday's context, knows the PR you were on
remembers you prefer small commits, picks up mid-thought
Session 47 · learned your deploy flow as a muscle
wrote itself a protocol for your test conventions
has scripts it built to catch the mistakes it used to make
you never taught it — it taught itself
The difference between an AI tool and an AI partner is memory.
Not RAG. Not vector search. Not "context window management."
Memory. The kind that changes how it works
because of what it's lived through with you.
◈ how it works
Soma's memory lives in .soma/ — a directory that grows alongside your project.
Everything is Markdown. You can read it, edit it, delete it. Nothing is hidden.
Heat controls what it remembers. Patterns it uses often stay hot — loaded every session.
Patterns it stops using cool off and fade. No configuration. Just use.
◈ AMPS — four layers of memory
Automations — triggered flows. Session starts, commits land, deploys fire.
Muscles — learned patterns. "How to ship code in this repo."
Crystallized from past sessions. The more it uses one, the hotter it stays.
Protocols — behavioral rules. "Always run tests before committing."
Shape how it works, not what it knows.
Scripts — tools it writes for itself.
Health checks, search utilities, verification passes. Built when it notices a recurring need.
◈ community hub
Browse community content at soma.gravicity.ai/hub.
Protocols, muscles, skills, templates — shared patterns from real workflows.
/install protocol breath-cycle
/install muscle docker-deploy
/list remote
BSL 1.1 © Curtis Mercier — open source 2027