Ship an agent into production and it handles most of the work — but a human still has to approve, route, or decide on the slice it can't. Pump Up is where that handoff lives.

It's the control plane for the team processing the residual slice of agent work: reviewing, approving, routing, handling exceptions. One coherent surface — queue → item view → audit log → manager view — with an immutable record of who decided what and why. The agent declares the work; we render it beautifully and keep the receipts.

We're agent-agnostic: run agents from any vendor or your own, and Pump Up ingests work from all of them.

• 📥 Queue — every pending decision in one real-time, filterable list, with priority and SLA timers.
• 🗂️ Item view — the decision surface, pre-loaded with all the context the agent gathered. Fast and keyboard-driven: approve, reject, edit-and-approve, or escalate.
• 📜 Audit log — an append-only history of every decision, always on. The record your compliance team will ask for.
• 📊 Manager view — throughput, SLA compliance, queue depth, override rates. Where the ops leader runs the team.

Drop a human in the loop in about 20 lines:

pip install pumpup-sdk      # Python
npm install pumpup-sdk      # TypeScript

const decision = await pumpup.requestApproval({
  type: "refund_over_threshold",
  context: { customer, amount, agentRecommendation },
  sla: "4h",
  routingHint: { team: "claims" },
});

The human decides in a fast, purpose-built UI; your agent gets the answer back.

→ Read the docs · SDKs & integrations (Python, TypeScript, OpenClaw — more by where developers build)

Building on an agent framework? Skip the wiring — first-class integrations live in pumpupai/pumpup, starting with OpenClaw.