Quickstart
Get an agent running durably on Nebula in about five minutes.
Install
npm install @nebula/runtimeDefine an agent
Wrap your existing loop. Declare tools and the guardrails that need a human.
import { agent, guard } from '@nebula/runtime'
export const checkout = agent({
tools: [lookupOrder, issueRefund],
guards: [guard('issueRefund', (a) => a.amount > 200)],
eval: './evals/checkout.suite.ts',
})Deploy
Push it. Nebula runs it durably, traces every step, and scales to zero when idle.
npx nebula deployChangelog
v0.9.0Jul 2026
- · Live trace replay against new models
- · Per-client tool scopes
v0.8.0Jun 2026
- · Evals in CI with per-case deltas
- · Mobile approvals
v0.7.0May 2026
- · Durable checkpointing GA
- · Audit log export