Give it a job, a way to check its own logic, and the tools to follow through then watch it run, live, with your whole team in the room.
No hidden logic every step in a run is one of these, sitting right on the canvas.
Where every run begins.
Runs a model against your instructions and the conversation so far.
Checks what just happened and picks the next path.
Calls a real endpoint and hands the response back to the workflow.
Marks where a run finishes and what it hands back.
Leaves context for the next person who opens the canvas.
Six node types. One canvas. No separate tools for building, testing, and running.
Connect an Agent node to an HTTP Request step and it can act, not just answer.
Condition nodes route the workflow based on what happened in the step before.
See everyone's cursor moving on the same canvas, in the same moment.
Chat with the workflow directly on the canvas — no separate staging step.
Every workspace, workflow, and run stays inside your org — nothing crosses over.
Swap the model running your agents without touching a single node.
A message comes in
Through the test panel, or wherever the workflow is triggered from.
The agent reads the graph
Every connected node's instructions become one system prompt, in order.
It can call a real step
An HTTP Request node becomes a tool the model can actually use.
You see the result
Streamed back in real time, with each tool call visible in the thread.
Multi-tenant by design every workflow is scoped to your organization at the database layer, not just hidden in the interface.
Every workspace, workflow, and run belongs to a single organization — nothing crosses over, ever.
Access is verified server-side on each call, not assumed from what the page shows you.
If something can't be confirmed as yours, you get a clear 'not found' — not a guess.
Start, Agent, Condition, HTTP Request, End — click any of them into place.
Draw edges between nodes to define what happens after each step.
Open the test panel and talk to the workflow exactly as a user would.
A workspace holds one canvas — its nodes, edges, and the graph you save. The Free plan includes one; Pro removes the limit.
You pick a label per Agent node; the underlying model can change without touching your workflow.
Yes — every workspace is a live, collaborative canvas. You'll see teammates' cursors moving as they work.
Yes, if it's connected to an HTTP Request node, an Agent can call it as a tool during a conversation.
The test panel locks with a clear upgrade prompt — your saved workflow itself is never affected.
One workspace, real agents, no card required.
