Cloud Action Agents
Action Agents run on ActionFlows servers. After you set one active, it can chat, use tools, and answer channels without you staying in Studio.
Cloud Action Agents
An Action Agent is the hosted AI assistant in ActionFlows. Cloud is not a second product. It describes where the agent runs: on ActionFlows servers, not on your laptop.
The product name is Action Agent. AI agent is the generic category (chatbots, coding assistants, local scripts). Do not treat "Cloud Agent" as a separate object in this product.
What "runs in the cloud" means
You design the agent in Action Agent Studio. ActionFlows then hosts the runtime.
After the agent is active, it can work when you are not in the app:
- A Slack, Discord, Teams, or email event starts or continues a session
- A Scheduled trigger ticks on an interval and starts a new session
- Your backend chats through the REST API, TypeScript SDK, or hosted MCP
- Tool calls, MCP, and Knowledge Base search run on ActionFlows during that turn
You do not keep a browser tab open, host a process, or run the model on your machine. Trigger.dev stays on ActionFlows servers.
Active vs Studio testing
Agents use the same active / passive idea as ActionFlows.
- Studio chat works while the agent is passive. Use it to test the prompt, tools, and knowledge.
- Channel triggers, the public API, the SDK, and hosted MCP need an active agent.
Set the agent active in Studio or on the agent Settings page. A channel binding also has to be ready (OAuth, mailbox, interval, and so on). See Triggers.
Action Agent vs ActionFlow
| Compare | ActionFlow | Action Agent |
|---|---|---|
| Shape | A graph of nodes | An ongoing conversation |
| How it starts | Start, schedule, API, or flow webhook | Studio chat, API, or a channel trigger |
| Lifetime | One run, then it ends | A session continues across messages |
| Human step | HITL pauses the graph | The agent is the conversational surface |
Use an ActionFlow when the work is a fixed pipeline. Use an Action Agent when the work is multi-turn chat with tools.
What you still configure
Cloud hosting does not mean the agent invents its own setup. You still pick:
- Model and system prompt (Model selection)
- Knowledge Base collections
- Skills
- MCP servers and tools
- Security policy
- Channel triggers
Secrets stay on named organization connections. The public API does not return prompt, tools, or MCP config.
Credits
Organization credits gate whether an agent turn starts, the same way they gate a flow run. Spend for one agent is on that agent's Usage page. See Usage & Cost.
Related
What is ActionFlows AI
An introduction to ActionFlows AI: the platform for building AI-powered automations visually with nodes, triggers, integrations, and models.
Node Anatomy
Understand the structure of an ActionFlows node: its inputs, outputs, handles, configuration, and how data flows through it on the canvas.