The agent orchestration platform with code intelligence and persistent memory built in, not bolted on, so every agent starts with the context it needs and every token drives real progress.
Reveals how your project fits together: every file, every function, and what calls what.
Preserves the decisions, fixes, conventions, and hard-won lessons that would otherwise be lost between sessions.
Together, they cut unnecessary exploration, reduce repeated work, and help agents spend their context and compute where it matters most.
Less rediscovery. Less wasted context. More progress per token.
Work with the agent turn by turn, hand a task to a crew of cheaper models, let an automation watch on a schedule, or accept a plan, set an optional dollar budget and leave it running.
Set up a reviewer, a test writer or a researcher, each with its own instructions and model. The lead decides who gets what.
See agents → CrewA frontier model leads, or none at all./crew-supervised: a leader plans, sets the check and judges each result. /crew-self-managed: one cheap or local model plans, codes and checks its own work.
See crew → AutomationsAutomate anything it can read.Use any model to watch your infrastructure, network, builds or logs on a schedule. When something needs attention, it starts an agent.
See automations → GoalsPlan it and leave it.Accept a plan, set an optional dollar budget, and the agent runs it in the background. Merge the result as one commit.
See goals →This one checks a Kubernetes cluster named pharma every 15 minutes. A small model on your own machine reads its nodes and pods and decides whether the cluster is healthy. If it is, nothing happens. If it isn't, an agent starts, investigates and reports what it found. Six automations come built in, and you can add as many as you want.
| Changes | Ask before each edit, or let them apply and keep or discard each file afterwards · control |
| Outside the project | On your machine, anything that reaches past the project asks you first |
| Undo | Every run is recorded before it starts. One click puts it back |
| Mid-run | Steer the agent while it works, or queue what's next |
| Where it runs | On your machine, or inside Docker or Podman · isolation |
| Your keys | API keys and sign-in tokens live in your system's keychain, never in a plain-text file or an environment variable |
Bring the subscriptions and API keys you already have, and mix them, with a different model for each agent if you like. Add tools from MCP servers, and let agents search the web or follow skills you write.
$10 a month after a 14-day trial · Questions