Big-AGI

Big-AGI is a multi-model AI workspace built by Enrico Ros and Token Fabrics. It runs in the browser as an installable web app, connects to AI services with the user's own API keys, and is also available as an MIT-licensed open-source project that can be self-hosted. Its signature feature, Beam, sends one prompt to several models in parallel and merges the replies.

What It Is

Big-AGI is a chat and workspace interface for working with many AI models in one place. Users add API keys for cloud providers or point it at local servers, then chat, attach files, draw images, use voice, and tune per-model parameters. The project states that it does not charge for model usage and that data is stored locally, with optional cloud sync on the hosted Pro tier.

How Beam and Merge work

Beam runs from 2 to 24 models in parallel during a chat, streaming each answer live. The models answer independently, so the user can see where they agree and where they diverge. A Merge step then combines the replies into one answer using built-in fusion modes (Fuse, Guided, Compare) or a custom merge prompt.

Control and transparency

Personas bundle instructions, attached documents and images, memory, a fixed model, parameters, and voice. The AI Inspector shows the exact request sent from the browser, including model, parameters, tokens, and cost. Other documented capabilities include split-pane chats, branching and forking, message editing and re-generation, diagrams (PlantUML/Mermaid), PDF extraction, voice calls, and automatic retries when providers fail.

Deployment options

The hosted app at big-agi.com works in the browser with BYO keys. The open-source branch can be self-hosted with Docker, deployed on Vercel, or run locally for development; the documentation covers configuration, optional services, and updates for self-hosted instances.