Skip to main content
alpha · honest about scope

Governed agents,
by construction.

A stateful, resumable agent-graph engine. Deterministic, checkpointed after every node, and gated for human approval — so an agent stops for a human before it acts.

Try in 5 minutes →Why Ailu?

Building with a coding agent? Start at /llms.txt and the For AI agents guide.

refunder.ts
import { createGraph, model } from "@ailu-ai/graph-sdk";

const app = createGraph({ name: "refunder" })
  .agentNode("decide", {
    model: model.anthropic("claude-sonnet-4-6"),
    prompt: { system: "Decide whether to refund the order." }
  })
  .humanGate("review")          // pause for a human before it acts
  .compile();

const run = await app.run({ request: "refund order #1024" });
// run.status === "suspended"  → stopped at the gate
await app.resume(run.runId);   // after a human approves
// status === "completed"
run()suspendedapprove (resolvedBy = you)completed

I want to…

evalEvaluate Ailu in 5 minutesbuildBuild a governed agentdeepShip a deep agentgateAdd an approval gatecomplyPass a compliance reviewmodelWire a model provideragentLet an AI agent author graphsshipDeploy to production

Pick your path

Evaluate in 5 minutes

Install, run a governed agent, watch it suspend at a gate and resume.

Start →

Build a deep agent

Plan with todos, spawn sub-agents, load skills — governed throughout.

Start →

Governance & compliance

Approval gates, attestation, determinism — end to end.

Start →

For AI coding agents

Machine-legible surface: /llms.txt, JSON Schema, a recovery loop.

Start →

Honest about scope: see what is stable today.