Skip to content
Open app Talk to sales

Let your AI agent work in your apps.

You set the goal and choose which apps, flows and files the agent can use. It decides the steps and carries them out. Anything you want to check first waits for your approval.

Start buildingTalk to sales
Process the invoice in my inbox.
Agent
Found the invoice· Maple · $4,200
Matched the PO· #PO-1183
Checked the budget· Under Q3
Payment · Maple Studio
Amount$4,200.00
Invoice#INV-2043
DueJul 30
Pay $4,200 to Maple?
Scheduled · logged in QuickBooks
Give the agent a task…

You choose its tools

Any app action, one of your own automations, an MCP server, or files you upload for it to read.

Your model, your key

An admin sets up the provider once: OpenAI, Anthropic, Gemini, Azure or Bedrock.

Approval before it acts

Put a person in front of any step that matters. A paused run keeps its place for 30 days.

How you decide what an agent can do.

Locked Inputs

Set the values the agent cannot change, and let it fill in the rest.

Your Own Model

OpenAI, Anthropic, Gemini, Azure, Bedrock, or any compatible endpoint.

External MCP

Let the agent use your own MCP servers too.

Human Approvals

The run waits for one person to approve. They need no account and no seat.

Runs That Resume

Finished steps are kept, and every tool call is written to the run log.

Custom Tools

Write your own integration in TypeScript.

Hosted Forms

Text, long text, file and toggle fields on a page we host for you.

Open Source

Agents work the same on our cloud and on your own instance.

Common questions.

An instruction, the tools you allow it, and the knowledge you give it. Its tools can be any integration, another of your automations, or your own MCP servers. Knowledge comes from files you upload or tables you keep current. Two limits: it remembers nothing between runs, and there is no live sync from SharePoint, Drive or Notion.

Yes. An admin adds the provider credentials once and every project can use them, so nobody else handles a key. OpenAI, Anthropic, Google AI, Azure OpenAI and AWS Bedrock work directly, and anything else connects through the OpenAI-compatible provider with a base URL and key. Admins also choose which models are available.

Three things: the run at 1 credit, each AI step at 2, 10 or 20 depending on the model tier, and each agent tool call at 1. A fifty-step automation with no AI in it costs 1 credit.

It gets cheaper rather than free. An AI step costs 1 credit with your own key, and you pay your provider for the tokens. Calling your own AI service through a custom integration or an HTTP step is not counted as an AI step at all.

Yes. Turn on the project's MCP server in settings and point Claude, Cursor or your own agent at the connection string. It creates and edits automations directly in your project, with nothing to import. It is on paid plans, and it builds them well but not perfectly.