Nodes
Every step you can put in an agent: AI, data, flow and app actions, and why each exists.
Nodes are the steps of an agent. You connect them on the canvas and data flows along the edges. Every step that runs is one Execution; see Plans and billing.
The starting node is covered in Triggers, and app actions in Integrations.
AI
| Node | Why |
|---|---|
| AI Agent | A model with tools, looping until it has an answer. See AI agents. |
| LLM | One model call, no tools: prompt in, text out. |
| Lucid | Ask Lucid about your own agents and data and pass the answer on, investigate an alert into issues, or improve an eval. Runs on Lucid credit, or on your own model keys. |
| Hand to coding agent | Give an issue to your coding agent (Cursor or Claude Code), by default the one the step before filed. It pushes a fix branch and reports back; it never opens or merges a PR. With no issue, or one already being fixed, the step is skipped with the reason. |
Data
| Node | Why |
|---|---|
| Query data | Read your events, runs, spans or users with a filter, a time window and a limit. |
| SQL | A read-only SELECT over your data, scoped to your team. Anything else is rejected. |
| HTTP request | Call any HTTP endpoint with a method, headers and body. |
| Python | Run Python in a sandbox: inputs in, outputs out, with pip requirements and a timeout. |
Flow
| Node | Why |
|---|---|
| Set fields | Build a new object from what came before. |
| Filter | Continue only when the condition holds. |
| If | Branch on a condition: true one way, false the other. A branch not taken never runs. |
| Loop | Turn a value into a list of items for the next step. |
| Wait | Pause before the next step. |
| Stop | End the run here, as a success or a failure. |
Apps
Slack, GitHub and Linear actions, covered in Integrations.
Next
- AI agents for the agent node in depth
- Integrations for app actions
- Triggers