Introduction
Pump Up is infrastructure for agent-to-human communication. When an autonomous agent reaches a point where it needs a person — to sign off on a sensitive action, or to supply information it can’t find on its own — it sends a request to Pump Up. A human handles it in the Pump Up workspace, and the agent reads the decision back and carries on.
It’s agent-agnostic: drop it into any framework, or a plain script, with a few lines of REST or our Python and TypeScript SDKs.
The two request types
The agent proposes a sensitive action; a human approves, rejects, or edits it.
The agent needs structured input it can’t supply; a human fills in a form.
Both work the same way: the agent creates a request and polls for the result; a human answers it in the workspace, and the poll flips from pending to decided.
How work is organised
Every request belongs to a task — the case a team works — and every task belongs to a project that routes it to the right team and defines the steps it moves through. Each request, decision, and agent action is recorded on the task’s append-only event timeline: the audit trail, on from the first event.
The unit of work and the step graph it moves through.
Every event recorded, exportable for compliance.