A visual AI workflow builder that compiles a React Flow canvas into a real LangGraph state machine at runtime.

Drag nodes onto a canvas, wire them up, and hit run — each node lights up live over WebSocket as it executes. Runs durably on a Redis-backed queue so a server restart mid-run doesn't lose work. Every published workflow is automatically registered as an MCP tool, so Claude Desktop can run it by name.

Includes 14 node types: LLM (Claude via LiteLLM with fallback), web search, sandboxed Python code execution, HTTP with SSRF guard, email, condition/switch branching, human-approve with signed email links, sub-workflow, loop, merge, delay, Jinja transform, vector memory, and trigger (cron/webhook/MCP).

Also ships credentials (AES-256-GCM encrypted, OAuth for Google/GitHub/Slack), team workspaces with RBAC, per-run cost tracking, monthly budget caps, analytics dashboard, and AI-powered workflow generation from plain English.

Stack used