Sigil
Read the docs Start free trial

The control plane for
AI agents in production.

Deploy, route and monitor every agent from one place. Sigil traces each step, enforces budget limits, and reroutes automatically when a step fails.

Start free trial Read the docs
$ npm install @sigil/sdk
acme-support/Runs
productionus-east-1
RECENT RUNSlive

support-triage

Step

    One place for everything your agents do after they ship.

    Each capability works on the same trace, so a failed step, its fallback and its cost are always the same record.

    Agents · acme-support · production4 deployed
    AGENTVERSIONTRAFFICP95
    support-triageclassify → retrieve → draftv14100%3.8 s
    refund-agentcanary rolloutv7 → v880 / 20%5.1 s
    invoice-extractorbatch, nightlyv22100%11.4 s
    research-digestscheduled 07:00 UTCv3100%48.2 s

    Wrap the agent you already have.

    No rewrite. The SDK sits around your existing calls and reports to Sigil.

    1. 1
      Install the SDKnpm install @sigil/sdk
    2. 2
      Define steps and fallbacksName each step and say where it reroutes on failure.
    3. 3
      Set a budget and deployRuns appear in the console with full traces.
    import { Sigil } from "@sigil/sdk";

const sigil = new Sigil({ apiKey: process.env.SIGIL_KEY });

export const triage = sigil.agent("support-triage", {
  budget: { daily: 60, alertAt: 0.9 },
});

triage.step("classify_intent", {
  model: "primary",
  timeout: "30s",
  // on timeout or error, rerun this step on the fallback
  fallback: { model: "fallback", retries: 1 },
  run: async (ticket, llm) => llm.classify(ticket.body),
});

    Put your agents behind one control plane.

    Start with one agent, then add the rest as you go.

    Start free trial Read the docs