Design AI workflows visually

Build ActionFlows on a canvas. Connect models, tools, and logic, then run them from the playground, a schedule, a webhook, or the API.

ActionFlow

An ActionFlow is the work you already do, drawn as a graph you can see. Start, think, connect, decide, deliver. Product, operations, and engineering sit on the same canvas. You do not wait for extra connector code.

The playground is the product

You drag nodes for models, the apps your team already uses, and the logic in between. If, loop, merge, wait. When a person still has to say yes, the ActionFlow pauses for approval on Slack, email, Teams, Discord, or a manual gate. Preview as you go. ActionFlows checks the graph before a run spends credits.

AI on the path, not beside it

Generate text, images, speech, structured objects, and files. Transcribe audio. Mix providers from the same organization credentials you already connected. The model is a step, the same way Slack is a step.

How the work actually starts

Start a run in the playground when you want to watch. Put a Start Scheduled node on the clock. Catch an inbound webhook. Call the runs API from your own product. Nest a subflow when a piece of the graph should stay reusable. Every run is stored in history with inputs, outputs, cost, and artifacts. You can see what happened. You can try again.

A faster start when you want one

Start blank, open a template, or clone a public listing from Explore. The graph becomes yours. Credentials stay on the organization, so every ActionFlow can reuse them.

From idea to a running ActionFlow

A few steps on the canvas. Then your team or your product can run it.

Create your ActionFlow

Draw the graph on the canvas. No code required.
Add models, apps, conditions, loops, and approval steps.
Connect outputs from one node to inputs on the next.

Connect models and tools

Use organization credentials for AI providers and apps.
Chain multiple models and integrations in one run.
Attach MCP servers when you need tools that are not a catalog node.

Run it

Start a run in the playground when you want to watch.
Start from an inbound webhook or from the runs API.
Nest a subflow when a piece of the graph should stay reusable.

Schedule tasks

Recurring: every N minutes, hours, or days, or a weekly time, in your timezone.
One-time: a date and time for the run that should not happen twice.
Each tick is a run in the same history as a playground run, with cost and artifacts.

Integrate with your app

Start runs from REST, the TypeScript SDK, or hosted MCP.
Send inputs and read outputs from your own backend.
Inspect, cancel, and follow run logs without opening the canvas.

Frequently asked questions

Start building AI workflows

Create a free account, open a template or a blank canvas, and run your first ActionFlow.

Get started Templates