storm

Agent Standard

STORM is an open, neutral standard for building autonomous agents. It collapses the gap between signal and action — letting any service ingest data, score decisions, and execute on-chain or off-chain with a single primitive. STORM exists to make agentic software boring, composable, and honest.

→ Spin up an autonomous agent in one line
app.use(
  stormAgent(
    {
      signals: ["http", "webhook", "stream"],
      policy: "score >= 0.7",     // Fire when confidence clears the bar
      sizing: "adaptive",         // Confidence-aware action commitment
      memory: "persistent",       // Cycles are logged and replayable
    },
  )
);
That's it. Drop in a single block and your agent listens, scores, and acts. Every cycle is logged and replayable — no orchestrator required.

Last 30 days

12.4K
Cycles run
184K
Actions fired
47
Agents deployed
99.2%
Uptime
Powered by
OPENAIANTHROPICGEMINIWEBHOOKSHTTPPOSTGRESREDISS3

How it works

A standard for turning any signal into autonomous action. Open source, MIT licensed, runtime-agnostic.

01

Ingest signals

STORM listens to any source — HTTP, webhooks, streams, model outputs — and normalises them into a single typed event bus.

02

Score the decision

Each event is scored against your policy. When confidence clears the threshold, the action is queued for execution.

03

Execute & log

Actions fire through your handlers — on-chain, API, or shell — with adaptive sizing. Every cycle is logged and replayable.

Your signal. Your agent. One line.

A real STORM swarm is running right now — four specialist agents deliberating in parallel, a coordinator returning a structured verdict. Watch it work.