Build an agent

Beyond the default assistant, you can stand up your own agents — each with a persona, its own model, and one or more triggers. They run under your identity with your tools, and record every run so you can see what they did.

Create one

From chat, describe the agent and how it should fire:

"Create an agent called 'Morning brief' — every weekday at 08:00, summarise my calendar and unread mail and send it to my Telegram."

Under the hood this creates an agents row (persona + model) and a schedule trigger. You can also manage agents from the Agents screen in the web app.

Triggers

• schedule — a time or cron-like window ("mon,wed 14:00", "every 30 minutes").
• sensor — fires on new events (schema in place; dispatch is maturing).
• webhook — fires on an inbound call.
• response / decision gate / agent-to-agent — react to an answered escalation, pause for a decision, or delegate to another agent.

Wire an agent org

Agents can relate to each other — delegate work down, escalate findings up — and you see the graph on a live org chart. When an agent is unsure, it lands an item in your escalations inbox instead of guessing.

A planning agent that reads your journal

A common pattern: a weekly agent whose persona calls journal_query and drafts content.

"Create a weekly agent 'Blog planner' — every Friday, read my eidan journal entries from the past week and draft two blog-post outlines."

This composes two primitives — the journal and scheduled agents — with no new code. That's the point: capabilities that call each other.