AI Models
Agents & Skills

Hire intelligence that gets it done.

Describe the outcome. Your agent plans the steps, uses the right skills, remembers what matters and reports back — while you stay firmly in control.

Start free Explore models
Agent run · live

  1. Understand

    Intent & context

    Done

  2. Plan

    Break into steps

    Done

  3. Act

    Run skills & tools

    Running

  4. Deliver

    Verified result

    Queued
Why agents

From answers to outcomes

Chat answers questions. Agents finish work.

  • Goal-driven

    Give a goal, not a script. Agents plan, adapt and recover on their own.

  • Skill-powered

    Search, read, write, calculate and call your tools through composable skills.

  • Memory-grounded

    Agents learn your preferences, projects and vocabulary over time.

  • Guardrailed

    Approval steps, scoped permissions and full run logs on every action.

  • Always on

    Bring a specialist into any conversation in one click.

  • Team-ready

    Share agents across your workspace with roles and audit trails.

Built in code or clicks

An agent in a dozen lines

Compose agents visually in the app or declaratively through the API. Same power, your choice of interface.

  • Pick skills from the library or bring your own
  • Choose memory scope: user, team or workspace
  • Require approval before sensitive actions
request.ts
const agent = await client.agents.create({
  name: 'Research Analyst',
  goal: 'Brief me on any company in under 2 minutes',
  skills: ['web.search', 'docs.read', 'sheets.write'],
  memory: { scope: 'workspace' },
  guardrails: { approveBefore: ['email.send'] },
});

await agent.run({ input: 'Brief me on our top 3 competitors' });
Skill library

Skills your agents can use today

  • Research

    Web search & read

    Find, read and cite sources, then summarise what matters.

  • Documents

    Docs & files

    Parse PDFs, spreadsheets and decks; extract, compare and draft.

  • Code

    Code & data

    Write, explain and transform code; run structured data analysis.

  • Comms

    Messages & email

    Draft replies and updates, with approval before anything is sent.

  • Integrations

    HTTP & webhooks

    Call every model from your own stack through one compatible API.

  • Custom

    Your own skills

    Wrap any internal capability as a skill and reuse it everywhere.

How an agent works

Understand. Plan. Act. Deliver.

  1. 01

    Understand

    Reads the goal, context and memory to grasp what “done” means.

  2. 02

    Plan

    Breaks the goal into steps and chooses skills and models for each.

  3. 03

    Act

    Executes, checks its own work and adapts when something changes.

  4. 04

    Deliver

    Returns a verified result with a full, reviewable trail.

Our principle
“Autonomy is only valuable when it is accountable. Every agent action is visible, reversible where possible, and yours to approve.”
The AI Models agent charter
FAQ

Agent questions, answered

How many agents can I run?

Three on Free, unlimited on Pro and above. Usage is measured in tokens, so you only pay for the work agents actually do.

Can agents take actions in my tools?

Yes, through skills you grant. Sensitive actions can require human approval before they run.

What happens if an agent gets stuck?

Agents retry, re-plan or ask you for help. Every run is logged so you can see exactly what happened.

Can I build custom skills?

Yes. Wrap any HTTP endpoint or internal function as a skill and share it across your workspace.

Your first agent is free

Create one in minutes. Watch it plan, act and deliver.

Start free Talk to sales