AI Agents
Configure OpenAI-powered agents for inbound WhatsApp, workflows, knowledge lookup, and appointments.
Overview
AI agents use your own OpenAI API key (Settings → AI or drawer → AI agents → credentials). Each agent defines:
- Main system prompt
- Primary and fallback model
- Context toggles (tags, pipeline, message history, local memory)
- Built-in tools:
- Tags and pipeline moves
- Escalate to human in inbox
- Knowledge table lookup
- Appointment tools (check availability, list / book / reschedule / cancel on allowlisted booking calendars)
- Optional inbound WhatsApp reply (always-on; pauses when the conversation is escalated in inbox)
Publishing and running agents typically requires a Pro/Enterprise plan (or an active trial), depending on your workspace billing.
Quick start
- AI credentials — save and test your OpenAI key.
- AI agents — create an agent, publish, enable inbound reply if needed.
- Optional: attach Knowledge tables on the Knowledge step.
- Optional: enable Appointment tools, allowlist one or more published booking calendars, set access to Read or Read & write, and choose a default calendar.
Appointment scheduling with AI
Agents do not book arbitrary Google calendars. They only use booking calendars you allowlist, and writes require the calendar to be published (active). Tools act for the current conversation contact.
| Tool capability | Access needed |
|---|---|
| Check availability | Read or Read & write |
| List contact appointments | Allowlisted calendar |
| Book / reschedule / cancel | Read & write |
Typical prompt outcome: offer slots → book → share confirmation (and Meet link when the calendar creates one).
Workflows
- AI agent node — runs a saved agent with tools; output in
{{ai.lastReply}}. - AI process node — one-shot personalization; output in
{{ai.output}}.
Inbound reply auto-creates a workflow: Message received → AI agent → Send WhatsApp reply.