Product

Everything a run needs to be trusted.

Five capabilities that turn a prompt chain into infrastructure you can put on call.

Run Graph

Watch every step, branch, and retry as it happens, with the exact inputs and outputs on each node.

const run = await agent("triage").start(ticket)
run.on("step", (s) => console.log(s.id, s.status))

Per-tool exponential backoff, declared in one file instead of scattered through your handlers.

retry: {
  "kb.search": { max: 3, backoff: "exponential" },
  "stripe.refund": { max: 1, requireApproval: true }
}

Every prompt, tool call, and token — retained and replayable, exportable as compliance evidence.

await runs.export({
  from: "2026-01-01",
  to: "2026-03-31",
  format: "jsonl"
})

Pause a run, review the state, approve or amend, and resume from the same step.

.step("stripe.refund")
.approval({
  when: (ctx) => ctx.amount > 100_00,
  notify: "#billing-approvals"
})

A hard ceiling per run. The agent stops at the limit instead of quietly overspending.

budget: {
  perRun: { usd: 0.40 },
  alertAt: 0.8,
  onExceed: "stop"
}