Product agents
Create, deploy, and operate workflow-backed AI agents from the Agents hub
Agents are a product layer on top of workflows: each agent is one workflow with a single Agent block, deploy surfaces (web chat, webhooks), tools, and skills. You manage agents from Agents in the workspace sidebar without opening the canvas for day-to-day edits.
What is an agent?
| Concept | Detail |
|---|---|
| Identity | agentId is the same as workflowId |
| Graph | Canonical template: trigger → Agent block (optional downstream blocks) |
| Metadata | Product fields live on workflow.variables.__aptlystarAgent |
| Runtime | Same executor and AgentBlockHandler as canvas workflows |
Create and operate
- Open Agents → Create agent and complete the wizard (basics, behavior, tools, optional deploy).
- Land on Agent detail with Overview, Channels, and Settings tabs.
- Deploy web chat from Channels or the deploy modal on the canvas.
- Use Edit in canvas when you need loops, extra blocks, or advanced wiring.
For complex graphs, keep using the workflow editor. The Agents UI is optimized for persona, tools, skills, model, memory, and channel setup.
Tools, skills, and cost
- Tools — Each selected integration operation sends its full JSON schema to the model on every turn. Prefer a focused set; the UI warns when you attach many operations.
- Skills — Names and descriptions appear in the system prompt; full playbooks load via
load_skillwhen needed. See Agent Skills. - Orchestrator mode — For large tool surfaces, switch to orchestrator mode so the parent agent delegates focused sub-sessions instead of loading every schema each turn. Configure limits under Settings → Advanced.
Channels
| Channel | Status |
|---|---|
| Deployed web chat | Available from Channels and the standard deploy modal |
| Slack | Configure from the agent Channels tab — see Connect Slack |
| Other messaging | Use workflow triggers on the canvas (Telegram, etc.) until dedicated agent channel UX ships |