Build a distributed system
by describing it in English.

OfficeSpeak turns a plain-English description into a small, always-on team of software workers — an office — that passes messages, keeps state, and reacts to the world. No programming needed to try it. Runs on the DisSysLab engine.

This is not a demo of one example

Real, verified groundwork already exists — the walkthrough you'll try is one story through a much bigger, working system.

33
example offices, running today
24 gallery apps + 9 direct examples — real code, actually run
20 + 500
sources & sinks
20 registered out of the box, plus 500+ MCP servers reachable beyond that
100%
crash-recoverable
real global snapshots — no lost or duplicate messages on resume
English
for building and debugging
describe it, correct it, and later ask it to explain a checkpoint or a run — all in plain English

A working gallery, not a single example

A sample of what's already built and running in DisSysLab's gallery — see FEATURES.md for the full, categorized list.

Built-in distributed-systems guarantees

The properties every generated office gets for free — never specified by the person describing it, never seen as implementation detail.

🔁

Distributed termination detection

Correct even with feedback loops in the agent network — not a fixed reply-count cap.

📸

Consistent checkpoint & resume

A genuine global-snapshot consistent cut across every concurrently running agent, including messages caught mid-flight — dsl run --snapshot-interval N.

🕵️

Explain a run, in English

dsl explain-trace merges every agent's activity log, ordered by a physical-time-grounded logical clock, into one causal story — narrated by Claude, not a template.

🗂️

Explain a checkpoint, in English

dsl show-checkpoint — what every worker's memory held, and what messages were still in transit, at one instant.

🧪

Isolated worker testing

Test one computational worker alone on hand-picked inputs to localize a bug — no concurrency in the picture.

Two stages, and it's normal to stop after the first

Stage 1

Describe it

No installation, no Python, no terminal. A conversation with Claude on claude.ai builds your office, explains it back — including what it assumed — and lets you correct it in plain English.

▶ Watch the 2-minute tour
Stage 2

Run it for real

Turns your description into an actually-running system, with real numbers. Needs a computer with Python, or someone who has one — right now, that can just be us.

▶ Watch the 2-minute tour