# Durable Streams

> The data primitive for the agent loop

Persistent, addressable, real‑time streams for building resilient, collaborative agentic systems.

## Table of Contents

### Docs

- [Quickstart](/quickstart.md): Get started with Durable Streams by following the Quickstart guide.
- [Core concepts](/concepts.md): Key concepts behind the Durable Streams protocol: streams, offsets, content types, producers, consumers, live modes, stream lifecycle, and CDN caching.

### Usage

- [CLI](/cli.md): Command-line tool for creating, writing to, reading from, and managing Durable Streams. Supports piped input, JSON mode, batch writes, and token authentication.
- [JSON mode](/json-mode.md): Stream structured data using JSON messages over a Durable Stream.
- [Fork](/fork.md): Create a new stream that branches from a source stream at a specific offset, inheriting its data without copying.
- [Durable Proxy](/durable-proxy.md): Put durable, resumable streaming in front of existing AI streaming APIs. Persists upstream responses into Durable Streams so clients can reconnect and resume reads.
- [Durable State](/durable-state.md): Structured state changes on top of Durable Streams. Work with typed insert, update, and delete events using MaterializedState, StreamDB, and Durable Sessions.
- [StreamDB](/stream-db.md): Type-safe reactive database in a durable stream. Define schemas with StandardSchema, query with TanStack DB, and use optimistic actions on top of Durable State.
- [StreamFS](/stream-fs.md): Shared filesystem for AI agents built on Durable Streams. Provides files, directories, metadata, content streams, and watch-based synchronization across agents.

#### Clients

- [TypeScript client](/typescript-client.md): TypeScript client for Durable Streams. Fetch-like stream() for reads and IdempotentProducer for exactly-once writes with batching and retries.
- [Python client](/python-client.md): Python client for Durable Streams with sync and async APIs. Generator-based stream consumption and IdempotentProducer for exactly-once writes.
- [Clients](/clients.md): Durable Streams client libraries for a wide range of languages, including TypeScript, Python, Go, Elixir, .NET, Swift, PHP, Java, Rust, and Ruby.

### Integrations

- [TanStack AI](/tanstack-ai.md): TanStack AI connection adapter for Durable Streams. Make chat sessions resumable, resilient, and shareable across tabs, devices, users and agents.
- [Vercel AI SDK](/vercel-ai-sdk.md): Vercel AI SDK Transport that makes chat sessions resumable, resilient, and shareable across tabs, devices, users and agents.
- [Yjs](/yjs.md): Sync Yjs CRDT documents over HTTP with automatic server-side compaction, snapshot management, and optional awareness (presence) support.

### Reference

- [Deployment](/deployment.md): How to deploy Durable Streams, including self-hosting with Caddy and running in production with Electric Cloud.
- [Building a client](/building-a-client.md): Use the conformance tests to one-shot a Durable Streams client in any language.
- [Building a server](/building-a-server.md): Use the conformance tests to one-shot a Durable Streams server. Covers protocol requirements, storage layer design and reference implementations.
- [Benchmarking](/benchmarking.md): Benchmarks package to measure server performance. Use it to validate that your server meets performance expectations or to compare configurations.