Small software / Operational agents
Turn a recurring agent task into software people can supervise.
Run model-driven work on a schedule, webhook, or user request. Keep its state, limit its tools, pause sensitive actions for approval, and expose every result and cost in the app.
Workers run deterministic code. Operational agents reason with tools, preserve state, and can pause for a person.
One example
Customer risk operator
Every Monday it reads CRM, billing, support, and product usage, explains new risks, prepares owner actions, and pauses before anything is sent.
Agent run / Monday reviewCustomer risk operator
Load 42 accounts · done
Explain 7 risks · done
Approve 3 drafts · waiting
LoopSwitchboardTablesDoor
How it comes together
Grant only needed tools
→Pause sensitive actions
→Review history, output, cost
What changes
A chat task becomes a bounded, inspectable operation.
01Run on schedules, webhooks, or user requests
02Preserve state and resume after failure
03Bind sensitive actions to human approval
04Expose history, cost, evidence, and outputs in a real UI
The cloud underneath
Use the pieces the app needs. They already work together.
Private alpha