Agento for developers

Build agents in code. Ship them with guardrails.

A typed SDK and REST API for agents that read, decide, and act in your stack, with approvals, retries, and audit logs built in.

Get an API key Read the docs

$npm i @agento/sdk

agent.ts

import { Agento } from '@agento/sdk';

const agento = new Agento({ apiKey: process.env.AGENTO_KEY });

const agent = await agento.agents.create({
  name: 'Lead researcher',
  instructions: 'Score new leads against our ICP and explain why.',
  tools: ['hubspot.read', 'gmail.draft'],
  approvals: ['gmail.send'],
});

await agento.runs.create({ agent: agent.id });

run_8f3a2c · Lead researcher

Read 48 new leads from HubSpot1.2s
Researched companies and scored fit38s
Drafted 12 intros in Gmail9s
Waiting for approval: send 12 emailsnow

API

Everything an agent needs, one API

Durable runs, scoped tools, and human approvals are primitives, not projects.

Agents as code
Define instructions, tools, and limits in TypeScript or Python, and version them with your app.
agents.create()
Durable runs
Every step is checkpointed, retried, and resumable, even across deploys.
runs.create()
Approvals
Mark any tool call as approval-required and resume when a person says yes.
approvals: ['gmail.send']
Webhooks
Get signed events for every step, approval, and result.
run.step.completed
Scoped keys
Keys per environment and per agent, with record-level permissions.
scope: agents:read
Audit log API
Stream every action to your SIEM or data warehouse.
GET /v1/audit