Giselle is an open source tool for designing and running AI workflows using a visual node-based editor. Apache 2.0 licensed.

Key features:

- Drag-and-drop workflow builder for chaining AI apps

- Multi-model support (OpenAI, Anthropic, Google Gemini) in one canvas

- GitHub integration for automating issues, PRs, and code reviews

- Built-in knowledge store with vector search

- Self-hostable with Docker or run locally

We built this because we needed a way to compose multiple AI models for complex tasks without writing glue code. The visual editor makes it easy to see the flow and debug when things go wrong.

Tech stack:

- Frontend: Next.js, TypeScript, Tailwind CSS

- Visual Node Editor: React Flow, Zustand, Tiptap

- AI: Vercel AI SDK, AI Gateway

- Database: PostgreSQL with Drizzle ORM

- Auth & Storage: Supabase

- Background Jobs: Trigger.dev (for long-running agent tasks)

- Analytics: PostHog

Happy to answer questions about the architecture or implementation.