Automator
Build, simulate, and run onchain workflows on a visual canvas.
Demo
About
Automator lets you connect contract calls, AI agents, forms, and logic on a canvas. You can build a flow by hand or ask the AI assistant to create and edit it. A flow can watch a token price and send an alert, collect a USDC payment through a mini-app, or query onchain data for an AI agent. Add screens to share it as an app, or define its inputs and return value to expose it through an API and MCP. You can simulate a flow, inspect each step’s output, and see where it fails. Onchain writes can be checked without sending transactions. Onchain transactions currently run on Base Sepolia and World Chain Sepolia.
How it's made
I used TypeScript, Bun workspaces, and Turborepo for the monorepo. The frontend is built with Next.js, React Flow, Zustand, Coss UI, and Tailwind CSS. The backend uses Elysia on Bun and PostgreSQL, with shared TypeBox schemas for API validation. The workflow engine is a separate TypeScript package. OpenAI and OpenRouter power the AI integration, which converts tool calls into graph edits. Custom JavaScript runs inside QuickJS with time and memory limits. The MCP server uses the official TypeScript SDK and shares the HTTP API’s execution code. I used viem for EVM calls and transactions, Privy’s React and Node SDKs for authentication and wallet signing, and World ID’s IDKit for identity verification. Subgraph queries go through The Graph gateway; balance and price triggers use The Graph Token API and Chainlink feeds.