Agent SDK
A small, readable API for giving your agent tools, context, and a place to work.
EXPLORETools, memory, and execution in one dependable harness.
From the first instruction to the finished work.
Research durable agent patterns. Turn the findings into a clear, useful brief.
A SHARED FOUNDATION FOR INDEPENDENT TEAMS
The brief becomes a plan.
The plan becomes something useful.
Start with an outcome, not a script. Give your agent the brief and the boundaries; it can work out the next useful step.
Different tasks. The same tools, context,
and room to get the work done.
Give the next decision a useful starting point.
Take a focused change from the brief to review.
Turn recurring questions into practical next steps.
Give each project its own files, memory, and permissions. Keep the right context close, and everything else out of the way.
We stopped teaching our agents how to keep running and started giving them better things to do.
Context is there. Tools work. The work gets done. We can finally spend our time on the idea, instead of the infrastructure around it.
Our agents have room to explore. We can still see every step, step in when it matters, and pick up the work without starting over.
Things we’re learning about building
agents that are good at the work.