Interleave
Interleave

Concurrency.Made visible.

Find the race. Replay every step. Understand why a fix works, with six interactive experiments and every schedule in a finite model.

Open the lab
Interleave on a MacBook, showing a completed lost-update execution in the installed macOS app.
6 interactive experiments
Every modeled schedule
Install on your desktop
Open source. Yours to explore.

Explore

Find the bug. Understand the fix.

Control two workers. Inspect every change. Then try a fix and put it through every possible order in the model.

Interleave running a lost update: two completed workers leave the shared counter at one.

Two increments. One lost update.

Step through a real execution and see exactly where shared state goes wrong.

Explore this execution

Actual captures from the installed Interleave app on macOS. Each view links to the same reproducible execution.

Experiments

Six ways for “it should work” to go wrong.

Familiar bugs, reduced to the moments that matter. Every experiment includes a broken implementation, a fix, and explicit assumptions.

Shared memory01 / 06

Lost update

When two increments become one.

How it breaksTime ↓
  1. Worker A: Read counter → 0
  2. Worker B: Read counter → 0
  3. Worker A: Write counter → 1
  4. Worker B: Write counter → 1
counter = 1Expected 2

Both workers start from the same value. The last write erases the first increment.

The fixMake the increment atomic.

Try this experiment

Investigate

Keep the whole investigation in one place.

01 / Follow the execution

Same code.
A different ending.

Rewind to any step and choose which worker goes next. See how a different order changes the result, then share a link that replays every decision.

Rewind a lost update
One counter. Two workers.
The failing ordercounter = 1
Rewind. Change the order.
A different ordercounter = 2
Illustrated lost-update schedules. Both start at zero.

02 / Understand the why

Your trace.
A conversation.

Ask the optional local AI tutor about the execution in front of you. It receives verified model context and generates its answer on your device.

WebGPU required. The first model download is approximately 1 GB.

How local AI works
From execution to explanation
Current executionWhy did one increment disappear?
Lost updateCounter: 1 of 2
Verified model context
Local AI tutorRuns in your browser with WebGPU
Illustrated assistant flow. You choose when to enable it.

03 / Keep the investigation

Leave yourself
a trail to follow.

Keep the replay and your notes together in a private investigation. Sign in to save your work and pick up where you left off, or export a Markdown report from the lab.

The public lab is always available without an account.

Open your workspace
Example investigation
LOST UPDATE / BUGGY MODEL

Two workers, one missing write.

ReplayA B A B A B

Both workers read zero before either writes. The second write replaces the first increment.

6 steps · Reproducible execution
An example of the context you can keep with your notes.

Install

A little lab.
A place on your desktop.

Install Interleave for its own window and a shortcut in your Dock or app launcher. The public lab works offline after it has loaded once.

Or keep going in your browser
macOS · Windows · LinuxNo installer download required

A progressive web app. Installation depends on your browser. Sign-in and cloud saves require internet. Local AI has separate hardware and model-download requirements.

Interleave installed on a MacBook, showing the lost-update experiment after both workers complete.
Interleave on macOS. A real execution in the installed app.

Questions

Before you
start exploring.

Read the model guide
Does Interleave run my production code?

No. Interleave is an educational workbench for finite, two-worker models. It does not scan repositories or prove arbitrary programs safe.

What does “every schedule” mean?

The engine explores every enabled terminal schedule within each built-in finite model. If a search budget is reached, results are marked partial. Counts are not production failure probabilities.

Is it free? Do I need an account?

The public lab is free, open source, and requires no account. Sign in only when you want to keep private investigations in the cloud.

Can I use it without an internet connection?

After opening the lab online once in the production app, its core experiments are cached on that device. Private saves and sign-in need internet. Clearing browser storage removes offline files.

Can I run it on my own infrastructure?

Yes. The MIT-licensed source includes the deterministic engine, UI, database migrations, tests, and deployment documentation. Follow the repository’s setup instructions to configure your own identity and storage.