The Intent Computer gives openclaw a memory that persists, skills that compose, and agents that run themselves.


Every agent starts from zero. It doesn't know what you decided yesterday, what you were debugging before lunch, or why you picked Postgres over SQLite. You re-explain. You repeat yourself. You lose momentum. The Intent Computer fixes this with a context graph that remembers, a skill graph that acts, and intent capsules that package everything an autonomous agent needs to run on its own.


The stack

04

Intent

What you want to happen.

capsule.yaml — a manifest that declares the goal, the skills, and the schedule. Everything else follows from this.

03

Knowledge

What the agent knows, who it is, and how it works.

SKILL.md files, identity.md, ~/Mind/ vault — methodology as natural language, identity as prose, memory as a filesystem-backed graph.

02

Engine

What makes it run.

The heartbeat scheduler, Agent Browser, Claude Code or Shelley — trigger evaluation, skill execution, autonomous decision-making between sessions.

01

Machine

Where it lives.

An exe.dev VM, your MacBook, a Linux server, a Docker container — persistent disk, a scheduler, and network access. The capsule doesn't care which.


The System

Context graphs, skill graphs, and the intent loop — how your agent gets persistent memory, composable capabilities, and autonomous decision-making.

/system →

Capsules

Portable seeds that deploy autonomous agents. A manifest, an identity, skills, and a vault — everything an agent needs to run on any substrate.

/capsules →

Declare the intent. The computer handles the rest.


Also compatible with

opencode · Claude Code · Pi Mono