Configuration
/ Where everything lives in the dashboard: knowledge, integrations, guardrails, Human Desk.
The tabs
Each agent lives under Dashboard, then Agents. Settings split into tabs, in the order you would set one up.
| Tab | What you set here |
|---|---|
| Profile | Name, personality, vertical, greeting, avatar |
| Knowledge | URLs, files, manual topics, live data connectors |
| Training | Automated test loops and topic coverage |
| Integrations | Widget embed, API key, hosted link, webhooks |
| Human Desk | Turn on handoff, set the escalation email and default assignee |
| Widget | Accent color, position, allowed domains |
Launch gate: 85% on all critical topics
Greeting & fallback messages
Each agent has two canned lines, customizable under Edit agent.
- Greeting: shown when the widget, hosted link, or React component opens. Defaults to a line matching the personality (Casual, Corporate, Efficient), works across every vertical.
- Fallback: used when the agent cannot answer from its knowledge base.
Leave either field blank to keep the default. API integrations can fetch the greeting with GET /api/v1/agents/{agentId}.
It picks the model for you
You do not choose a model per message. Humaner looks at how hard the question is, the escalation risk, and the vertical, then routes it.
- Haiku for the high-volume, easy stuff (order status, FAQs). Fast and cheap.
- Sonnet for the hard stuff (multi-step reasoning, sensitive topics).
- Smart response reuse for questions already answered well.
Guardrails come with the vertical
Each vertical ships with forbidden topics and escalation rules. A wellness agent refuses medical advice. A hospitality agent escalates safety issues right away. These run before a single word reaches the customer.