Y CombinatorBacked by Y Combinator

Streams as a cloud storage primitive

The serverless API for unlimited, durable, real-time streams.

s2-lite an open source S2 you can self-host

Used by

The missing
building block

S2 is to moving data what object storage is to static data.

Like if Kafka and S3 had a baby. It turns the log into an actual web resource.

Infinite streams

unlimited bottomless auto-create auto-cleanup

Streams get a first-class treatment, so you can have any number of them, per device, per user, per session, and so on.

Append quickly

<40ms
Express
<400ms
Standard

Append records atomically to the tail of a stream. Every record is durably ordered on object storage before being acknowledged.

Read from anywhere

Reactive
Follow live updates
Resumable
Pick up where you left off

Read efficiently from a sequence number, timestamp, or offset from the current tail. Guaranteed to be strongly consistent.

Simple, powerful API

Working with streaming data shouldn't feel exotic.

True web resources

Interact with S2 streams using REST APIs.

  • Append to and read from streams easily, whether from your backend or client-side code.
  • Long-running and real-time tailing reads, via SSE or HTTP/2 sessions.
  • SDKs available for TypeScript, Go, Rust, and Python.

Granular access controls

Issue access tokens so that clients can safely and directly access streams.

  • Just like streams, no limit on number of access tokens.
  • Scope by resources and the operations allowed on them.
  • As ephemeral or permanent as you need, with automatic expiration and support for revocation.

Convenient CLI

A command-line interface that makes streaming data as simple as local files.

  • Create and manage basins (like a bucket), streams, and access tokens.
  • Streaming append and read sessions to easily interact with streams.
  • Use it interactively with s2 -i.

Streams on tap

Check out the applications developers are powering with S2.

Project Board
A
Alice
B
Bob
C
Charlie
To Do
In Progress
Done
Sync Stream

Local-first and multiplayer experiences

Build stateful collaborative experiences reliably with S2. Real-time does not have to mean ephemeral. Try our multiplayer AI chat.

A1
Planning
A2
Research
A3
Execution
↓
Session Stream
Plan created
Data gathered
Task executed

Agent sessions

Event sourcing for agents, with a stream per session. Branching, multi-agent coordination? Now tractable.

Broadcasting...
42,851
events
Durable stream
842
concurrent readers

Real-time feeds

Broadcast data to unlimited subscribers. Let them consume from where they left off, be it seconds or days ago.

sboxer - browser-423af

Sandbox execution events

Stream per sandbox instance for observability. Your clients can access historical and real-time events without custom proxy infra.

Built for scale

An elastic system with a focus on correctness.

Unlimited streams
Model streams at the granularity that makes sense for your application. What you thought serverless meant.
Bottomless storage
Streams can grow indefinitely in volume. Data lives in object storage, keeping costs low.
Fast
Producer to consumer in <50ms (same region, p99). Easy to relax requirements and save.
High throughput
Each stream can be a trickle or a torrent — up to 100 MiBps writes. Massive read fanout? No problem.
Crafted in Rust
Efficient and memory-safe infrastructure, exhaustively tested using deterministic simulation.
Secure
SOC 2, HIPAA, and GDPR compliant. Best practices and organizational controls to keep your data safe.

Ready to start streaming?

Join developers counting on streams as a cloud storage primitive.

No credit card required · Your first stream in seconds