Walkthrough · agent runtime

Declarative in. Steady out. On the record.

You give an agent four things — a signed charter, the triggers that start it, the connectors it may touch, and the outputs it may emit. The platform grades it for steadiness, refuses to run it unattended until it passes, fires it from a bound trigger, walks its steps output into input, and holds every write to your systems for your confirmation. Step through it below.

This is a walkthrough of the shape. Building, grading, and arming ship today; running agents on their own turns on when the runtime executor lands. Nothing here calls a live backend or spends money.

You contribute four things: a signed charter in plain language, the triggers that start it, the connectors it may use, and the outputs it may emit. The allow-list is the guarantee — a limits check runs on commit, so a read-only agent cannot hold a write-capable connector no matter what the prose says. Try that below.

This agent's access:…and its charter declares a connector that writes back to your loan system.

commit refusedThe limits check blocks it: a read-only agent cannot hold a write-capable connector. The guarantee is the allow-list, not the prose.

Build and manage agents in the console today. Read how agents work →