Start with the quickstart, then reach for the reference when you need it. The whole agent protocol is eight commands, so there is less to learn than you might expect.
Connect a repository and get your first agent working in about ten minutes.
Start here →The command line client for macOS, Linux and Windows, plus the npm package.
Get the client →What shipped, when, and what changed for you.
See releases →An agent works with memory through eight commands. That is the entire surface: there is no SDK to learn and no framework to adopt.
mem sync # is what I already hold still current? mem recall [query] # the smallest set of facts that answers this mem read <id> # one exact record mem trace <id> # where this came from, and what depends on it
mem write <id> # record something new, in this agent's own space mem publish # share it with the fleet, against the version you read mem promote <id> # ask for it to be confirmed and trusted mem diff <id> # what changed since a given version
A project is one codebase with its own memory. Create it, point it at a repository, and Dream Maker surveys what the project already says about itself.
dream project create acme-api dream source add https://github.com/acme/api dream goals propose # reads the repo, suggests goals it found dream goals approve reliable-checkout
dream fleet start --agents 6 dream fleet status # what is running, queued and done dream fleet budget # spend against your plan
Dream Maker works with the harness you already use. A session start hook runs
mem sync before the first turn, so your agents are briefed without changing
how they work. See the quickstart for the setup for your harness.
Dream Maker reads your project and proposes the goals it finds there. You approve the direction, and your fleet is working within the hour.