# durable-streams/durable-streams

The data primitive for the agent loop.

Repository: https://github.com/durable-streams/durable-streams
Canonical: https://ross.abutalabs.com/products/durable-streams
Homepage: https://durablestreams.com
Language: TypeScript
License: MIT
License Family: permissive
Topics: agents, ai, data, durable, real-time, sessions, transport
Last push: 2026-08-12T17:58:17+00:00

## Health v2 (maintenance only)
Score: 74/100 (v2, computed 2026-09-03T02:39:23.370411+00:00)
- activity 97, release rhythm 75, longevity 19
- inputs: {"age_days": 272, "days_push": 21, "days_rel": 91, "gap_med": 0.0, "n_releases_24m": 115}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1674, forks 66 (observed 2026-08-28T04:05:20.287889+00:00)

## What it is
Durable Streams is a protocol and tooling for persistent, addressable, real-time data streams over HTTP, with offset-based resumability, catch-up reads, and live tailing. It targets resilient AI agent sessions and collaborative multi-user, multi-agent systems, with clients in TypeScript and Python plus integrations like TanStack AI, Vercel AI SDK, and Yjs.

## Use cases
- stream llm token responses with resume across reconnections
- build resilient agent sessions with replayable event streams
- stream tool outputs and progress events to clients
- sync database changes to web and mobile clients
- sync CRDTs for collaborative editing
- build event-sourced architectures with client-side replay
- tail streams live over sse with guaranteed delivery

## When to choose
- you need ordered, replayable, resumable streams over plain http
- you are building agentic or collaborative AI apps that must survive reconnects
- you want exactly-once delivery semantics to browsers, mobile, or native clients
- you want integrations with TanStack AI, Vercel AI SDK, or Yjs

## When to avoid
- you need a full message broker with queues, routing, and fan-out like kafka or rabbitmq
- your use case is simple request/response without durability or replay
- you need non-http transports or strict low-latency intra-cluster messaging

## Facets
- artifact type: library
- maturity: active
- function: streaming, http-server, http-client, message-queue, sdk, cli
- domain: microservices, developer-tools, apis, large-language-models
- platform: python, cross-platform, cli
- tags: durable-streams, agent-loop, resumable-streams, event-sourcing, collaborative-ai, offset-based-resumability, live-tailing, tanstack-ai, vercel-ai-sdk, yjs-provider, ai-agents, real-time, typescript, nodejs, web-server

## Member repositories
- durable-streams/durable-streams (main) score 74

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:20.287889+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:42:31.579296+00:00, confidence not recorded.
  - readme: https://github.com/durable-streams/durable-streams (fetched 2026-08-28T04:05:20.287889+00:00, sha 6c9b4f167c8c)
  - homepage: https://durablestreams.com (fetched 2026-08-29T11:15:40.059989+00:00, sha 546ac27049c7)
  - site_page: https://durablestreams.com/quickstart (fetched 2026-08-29T11:15:40.068914+00:00, sha 449bf0e6cd96)
- Data as of 2026-08-30T08:39:29.467469+00:00.