Agent Settings
Open from the Settings tab inside any agent.
General
- Agent name — shown in the dashboard and chat header
- Description — internal note about what this agent does
- Agent type — Knowledge (default) or Action. Knowledge agents are conversational chat experiences; Action agents run automated workflows on triggers. For action agents, trigger configuration lives in the Overview tab rather than in template settings.
AI behaviour
| Setting | What it does |
|---|---|
| Model | LLM provider and model — OpenAI GPT-4o, GPT-4o mini, Anthropic Claude |
| Temperature | Low = precise and consistent, High = more creative and varied |
| Response length | Concise, Balanced, or Detailed |
| System prompt | The core instruction that defines who the agent is and what it does |
| Supplemental instructions | Extra rules appended to the system prompt |
| Brand guidelines | Strict / Casual / Inherit from workspace |
Conversation Behaviour
Control how your agent handles follow-up suggestions, conversation resets, and new conversation creation.
| Setting | What it does | Default |
|---|---|---|
| AI Quick Replies | When no journey-defined buttons exist, the AI suggests 2–3 contextual follow-up options after each response. Journey-defined buttons always take priority. | On (Conversational), Off (Document Generator) |
| Allow "Start Over" | Users can say "start over" or "reset" to return to the beginning of the conversation journey. After 6+ messages, a greeting like "hi" also triggers a reset. Conversation history is preserved. | On |
| New Conversation Button | Shows a button in the chat header after 3+ messages that lets users start a completely fresh conversation. The previous conversation is saved to history. | On |
Escalation
Define when the agent should hand off to a human:
- Triggers — keywords or topics that should always escalate (e.g. "refund", "complaint")
- Behaviour — Offer to Connect, Auto Escalate, or Provide Contact Details
Safety & guardrails
| Setting | What it does |
|---|---|
| Competitor names | Agent redirects instead of comparing when these are mentioned |
| Forbidden topics | Topics the agent should never discuss |
| Negative keywords | Words the agent should never use |
| Deflection message | What the agent says when a forbidden topic comes up |
| Required disclaimers | Text that must appear when certain topics are mentioned |
Deploy
- Publish / Unpublish — go live or take offline
- Landing page URL — the public link to share
- Widget embed code — the
<script>tag for your website
See Deploying Your Agent for full instructions.
Groups
Coordinate this agent with others via Agent Groups:
- Join a group — select an existing group from the dropdown and click Join
- Create a group — enter a name and click Create to start a new group
- Leave a group — click the X button next to a group name
Agents in the same group share context. Knowledge agents automatically see recent run results from other agents in their groups, giving them awareness of what's happening across your automation. Use groups alongside the Trigger Agent journey node for full multi-agent orchestration.