Docs
Build on the runtime your agents answer to.
Octos is a governed runtime for agent compositions — it composes a task, makes the decision, grades it against the outcome, renders to any surface, and keeps memory true. These docs show you how.
Early docsNarrative pages — concepts and use cases — are live. Reference is being grounded against the runtime and publishes section by section as each part is verified.
The five verbs
Everything in Octos is one loop. Learn the five verbs and you know the shape of the whole system.
- Compose — a vertical is a configuration of services and checks; the engine assembles the right stack per task.
- Decide — the runtime makes a call under your checks, with the evidence attached.
- Grade — every decision is scored later against what actually happened.
- Render — the same composition surfaces as chat, a CLI, or a full app.
- Govern — decisions, evidence, and memory are kept as an audit-ready record.
Start where your task is
Three ways in, depending on what you're doing right now.
Ship
Quickstart
Register the door and make a governed write — end to end.
Map
Use cases
Find the surface closest to your problem — org, life, worlds, or your own.
Understand
Concepts
Services, the one door, state as a log, frames — the model underneath.