Durable Streams

Durable Streams is an open-source data primitive for the agent loop, providing persistent, addressable, real-time streams over HTTP. It enables resilient AI agent sessions, collaborative multi-user systems, and reliable client-side data delivery with offset-based resumability and exactly-once message delivery. Built on 1.5 years of production use at Electric for real-time Postgres sync, it reliably handles millions of state changes daily. The protocol is content-type agnostic, CDN-friendly, and works across web browsers, mobile apps, native clients, IoT devices, and edge workers.

• Offset-based resumability — clients resume from any point after disconnection, page refresh, or tab switch without losing data
• Real-time tailing — supports long-poll and SSE modes for live streaming with catch-up from any offset
• AI conversation streaming — stream LLM token responses with resume capability across reconnections using the Durable Proxy
• Agentic app support — stream tool outputs and progress events with replay and clean reconnect semantics
• Multi-client SDKs — official client libraries for TypeScript, Python, Go, Elixir, C#/.NET, Swift, PHP, Java, Rust, and Ruby
• CLI tooling — create, read, append to, and tail streams from the command line
• StreamDB & StreamFS — higher-level abstractions including a type-safe reactive database and a filesystem with real-time sync built on streams
• Durable State — sync structured state using insert/update/delete operations over streams
• TanStack AI & Vercel AI SDK integrations — drop-in transport layer for popular AI frameworks
• CDN-native scaling — HTTP-native design enables massive fan-out to millions of concurrent viewers from a single origin
• Production Caddy plugin — deploy a production-grade server using the official Caddy plugin
• MIT licensed — free to use, modify, and distribute; self-host with the Node.js reference server or Caddy plugin