Introducing Volt Agents 2.0: autonomy with guardrails

A year ago we shipped our first AI step. Teams loved it — and then immediately asked for two things: let it do more, and make sure it never does the wrong thing. Volt Agents 2.0 is our answer to both.

What's new

Agents can now break a goal into steps, choose which integrations to call, and loop until the job is done. Every action is recorded in the run log, so you can see exactly what happened and why.

• Planning — describe the outcome, the agent drafts the steps.
• Tool use — any of our 300+ integrations, plus your own SDK functions.
• Approvals — pause for a human whenever a rule says so.

Guardrails, not hope

Autonomy without limits is a liability. Each agent runs inside a policy you define in plain TypeScript:

agent('refunds', {
  tools: [stripe.refund, slack.post],
  policy: (action) => action.amount < 50 ? 'allow' : 'ask',
})

“Approvals were the unlock. Autonomy where it's safe, a human where it matters.” — Jonas Lindqvist, Halcyon

Available today

Agents 2.0 is live for every Pro and Scale workspace. Starter workspaces get 100 agent steps a month to try it out. Compare plans or talk to us about custom models.

---

Thanks to the 600 design partners who stress-tested agents over the summer. You made this better.