Connect your agent
Where Humaner Intelligence fits inside a reply loop you already own.
Who owns what
| Concern | Your agent | Humaner |
|---|---|---|
| Chat UI / widget | Yes | No |
| Model routing and prompts | Yes | No |
| Persona, tone, voice | Yes | No |
| Knowledge retrieval | No | Yes |
| Guardrail topic list | No | Yes (advisory) |
| Runbook matching | No | Yes |
| Tickets, routing, SLA | No | Yes |
| Visitor memory store | No | Yes |
Guardrails are advisory on Custom
Humaner never sees your prompt, so it cannot block a topic for you. The guardrail list is returned to your agent and you enforce it. Cloud enforces the same list inside its hosted prompt.
The loop
Hover to zoom
Where each call belongs
- Before answering anything factual: retrieve. Use the grounded flag to decide whether to answer or defer.
- When the question maps to a known procedure: match-runbook. The response names which of your dashboard runbooks matched — steer the reply toward that procedure.
- When the customer asks for a human, or retrieval is not grounded: create-ticket with the transcript.
- When you learn a durable fact about the visitor: remember, then memory on the next session.
Surfaces that stay Humaner-only
Widget, Link, and React embed the hosted Humaner agent, so they return 403 on Custom. Hosted chat at /api/v1/chat is closed for the same reason. If you want the hosted widget instead of your own agent, that is the Humaner plan.