giselles-ai/giselle

What it solves

Giselle is an open-source AI workflow studio for building and running AI agents locally or on your own servers. It was created to make human-AI collaboration seamless, giving teams a visual way to design, connect, and automate AI models for tasks like research, code review, and document generation — without relying on a hosted cloud service.

How it works

You run Giselle on your machine with just a few commands and add at least one API key from OpenAI, Anthropic, or Google AI. Once started, you use a drag-and-drop Visual Agent Builder to compose agents from multiple models (like GPT, Claude, and Gemini). The agents can pick the best model per task, and you can attach a Knowledge Store that supports GitHub vector storage for searching code and data. In development are team collaboration features and a template hub.

Who it's for

Developers and non-engineers who want to create custom AI agents and workflows for product delivery, research, or automation. It's especially suited for those who prefer self-hosting and local control over their AI tools. The included "Vibe Coding Guide" also helps people using AI coding assistants to build on Giselle without deep programming expertise.

Highlights

• Visual, drag-and-drop builder for creating and modifying AI agents.
• Multi-model composition: agents automatically select the best model (OpenAI, Anthropic, Google AI) for each task.
• Knowledge Store with GitHub vector store integration for searching code and data.
• Speed: get running locally in under 2 minutes.
• Open source (Apache 2.0) and self-hosted; cloud service discontinued.
• Use cases include research assistance, AI-powered code review, document generation, and multi-model workflow automation.