An agent's output is structured JSON. An integration is just a place to send it. Everything here is optional — the API and webhooks cover anything not listed.
Push triage results onto tickets as fields and internal notes.
Classify conversations and attach drafted replies for an agent to send.
Write lead scores, segments and reasoning onto contact records.
Qualification output as custom fields, with the reasoning stored on the lead.
Post run results, approvals and failures into a channel.
Pull source documents straight from a folder for extraction or research.
Watch a folder and run extraction on anything new that lands.
Write meeting summaries and research briefs back as pages.
Extracted fields and cleaned records written straight to a base.
Post the full run payload anywhere that accepts JSON.
Everything the dashboard does, available as an endpoint.
Trigger runs from 6,000 other applications without writing code.
Most of these are a thin layer over the same thing: take the run payload, map a few fields, write it to an API. If your tool is not on the list, a webhook and twenty lines of code gets you the same result, and you keep control of exactly what gets written where.
We are deliberately not building a deep two-way sync with anything. Bidirectional sync is where integrations go to become a support burden, and it is rarely what the work needs.
Each connection stores its own credential, scoped to the minimum permission the integration needs. Revoking a connection revokes the credential immediately; runs continue, they just stop delivering to that destination.