Workflow
A workflow is the thing you build on the canvas — a diagram of nodes that defines how an AI task runs from input to output. Each workflow belongs to your organization and can be edited, tested, and deployed.Node
A node is a single building block on the canvas. The platform provides four node types:| Node | Purpose |
|---|---|
| Input | Defines the information the workflow needs to run. |
| Agent | An AI step configured with instructions, a model, tools, and a knowledge base. |
| Output | The final result the workflow returns. |
| Group Orchestrator | Coordinates multiple agents working together as a team. |
Agent
An agent is the core working unit of a workflow — an AI step you configure with:- Instructions — the system prompt that defines its role and behavior
- Model — which AI model powers it
- Tools — external capabilities it can call
- Knowledge Base — content it can search
Run
A run is a single execution of a workflow. Every time you test in the Playground or trigger a workflow through the API, a new run is created. Each run has a status (pending, running, completed, failed, or cancelled) and a full trace of what happened.Thread
A thread is a conversation session. Multiple runs can belong to the same thread, which lets an agent carry context from one message to the next — like an ongoing chat rather than a one-off question.Deployment
A deployment is a published, versioned snapshot of a workflow. You edit freely on the canvas, but only a deployed version can be run by your team or called through the API. One deployment is active at a time — that’s the version that serves traffic.Model
A model is a configured AI model the platform can use to power agents. You add models under Models and select them when configuring an agent.Tool
A tool is an external capability — like sending a message or fetching data from another service — that an agent can use during a run. Some tools require a tool connection to a linked external account.Knowledge Base
A Knowledge Base is a collection of uploaded documents that the platform indexes so agents can search them and ground their answers in your own information.Memory
Memory lets a workflow retain context across the messages in a thread. It’s configured at the workflow level by connecting a memory provider.Guardrail
A guardrail is a workflow-level safety policy. The platform supports two types: PII detection and prompt-injection protection.Organization
Your organization is the workspace that contains all your workflows, resources, and members. Members are invited with a role that determines what they can do.Next steps
Build your first workflow
Put these concepts into practice with the Quickstart.