What is turbostream?

TurboStream is a Go based backend server designed to connect real-time data streams to AI language models, letting you query and analyze fast-moving data with natural-language questions. Think of it as a bridge: data flows in continuously, and you can ask an AI what is happening right now and get a live answer. The server exposes REST API endpoints and WebSocket connections, with WebSocket being a protocol that keeps a persistent, two-way connection open between client and server, so data and AI responses can arrive in real time without repeatedly opening new connections. It supports seven different AI providers out of the box, including Anthropic, OpenAI, Google Gemini, Mistral, xAI, and a local option via Ollama, using a bring your own model approach where you supply your own API key for whichever provider you prefer. Data is organized into feeds, which you can publish to or subscribe to through a built-in marketplace. Each feed can have its own isolated AI context through topic routing, so a feed tracking one asset does not bleed context into another. The server also uses a compact data format called TSLN that the README says reduces token usage by 40 to 60 percent, helping keep AI API costs down when analyzing high frequency streams. User accounts are protected with JWT-based authentication, a standard token system for secure logins, plus optional two-factor authentication. The project uses MongoDB as its database and can be deployed with Docker or on Railway. A companion terminal UI client is available in a separate repository, and a commercial web frontend is offered on top of the open-source backend. The full README is longer than what was shown.