Daloy workflow builder
A working AI workflow builder in the spirit of n8n and Zapier, shown on a made-up pottery studio in Bristol. You wire triggers, AI steps, approvals and app actions together on a canvas, then watch each run step through it live: an email gets sorted, a wholesale lead gets scored and logged, and the AI's reply waits for a person to approve it before it's sent.
- A schematic canvas editor: drag steps around, drag from a port to connect them, if/else and approval steps with yes and no branches, and a side panel to edit each step. Edits save as you go and every save is a new version
- Templates between steps: any field can use {{step.field}} from earlier steps, with click-to-insert chips and a warning when a field points at a step that hasn't run yet. Renaming a step rewrites every reference to it
- Four AI steps on Workers AI: extract fields with types, classify into your own labels with a reason, write an email in a set tone from the studio's real policies, and summarize with sentiment. Each one returns checked JSON
- Durable runs on Cloudflare Workflows: every step retries on its own, a Wait step sleeps, and Ask for approval pauses the run until someone approves, edits or rejects the draft, even days later
- A live run view: the path lights up on the canvas as steps finish, branches not taken fade out, and the run log shows exactly what went into and came out of every step
- Five triggers fed by sandbox apps: an inbox, the studio's wholesale form, invoices you can mark overdue, a schedule, and a real webhook URL you can hit with curl. Actions land in sandbox sheets, team chat, a CRM and sent mail, each linking back to the run that made it
Cloudflare Workers, Cloudflare Workflows, D1 (SQLite), Workers AI, TypeScript