What an Agent is

An Agent captures your method as a workflow: the inputs it needs, the steps it takes, the checks it applies and the result it delivers. Each run follows the version you published. For a research firm, an Agent can be the repeatable core of a research product.

The lifecycle

1. Teach. Describe the method in conversation. Axe drafts a workflow you can read and correct. See Teaching an Agent.
2. Test. Try the current draft on real inputs in a fresh Task, without changing what anyone else runs.
3. Publish. Make the draft the version future runs use. Axe checks the workflow first: problems must be fixed, and you review any warnings.
4. Run. Type / in any Task and pick the Agent. See Running an Agent.

Agents are included in an eligible trial and a paid plan. A successful credit purchase also unlocks them. Runs use credits from your organisation's balance. See Billing and credits.

Versions

Every run pins the exact version it started with. Later teaching or a new publication never changes work that is already queued, running, paused or finished. To change future runs, correct the draft and publish again.

Archiving an Agent stops new runs without deleting its versions or history.

Who can see an Agent

Agents are private to their owner. Organisation membership and billing never grant access to another member's method, sources or outputs. Each run rechecks current permissions, so a published version cannot carry access that has since been revoked.

Not yet available

Schedules, triggers, shared authoring, delegated publication, automatic distribution to your customers, nested Agents and a marketplace are outside the current product.