Build a system. Send a request. See what happens.
A visual system design workbench with executable diagrams, traffic experiments, and incident challenges.
Try it · Get started · Try a scenario · How it works · Development
Open the live app. No account and no API keys.
Liveboard turns a system diagram into a running simulation. Connect a client to a backend, add a cache and database, then follow a request through the system. Change the settings and measure the effect.
Use it to explore a design, practice an interview, or build an exercise for someone else.
- Open Liveboard. A new browser starts on the cache-aside template. Templates stay open in the Add panel, and an existing saved board is left as you left it.
- Send
GET /users/1. The backend misses the cache, reads the database, and stores the record. - Send it again. The second request is a cache hit.
- Open Run, then Traffic. One request per second fits the default backend. The slider says when a faster rate will queue before anything fails.
Pause with Space and scrub Replay to walk that request along the edges. After an interview experiment, choose Watch on the board to play the trace on the diagram. Press F to present. Copy a share link from the Board menu. The link is a compressed board in the URL hash.
- Build on a canvas. Add components, connect their ports, and configure each component or connection in the inspector.
- Follow execution. Inspect request traces, responses, cache entries, and queue state. Pause or step through the simulation.
- Test under load. Compare repeatable traffic experiments using success rates, response percentiles, database reads, and queue pressure.
- Introduce failures. Take components offline, slow database connections, or configure retries and circuit breakers.
- Solve incidents. Diagnose broken starter systems and verify a fix against traffic objectives and resource limits.
- Prepare an interview. Record requirements and design decisions, estimate capacity, and export an HTML report with measured evidence.
Your workspace saves in this browser. Two tabs coordinate through a storage lock and show a conflict when both have unsaved edits. Download boards as JSON, copy a share link, or drop a file onto the workspace. Exercises and submissions are portable files too. No account or API keys are required.
Use Node.js 22.x starting at 22.12, 24.x, or 26+.
npm ci
npm run devOpen localhost:5173.
A new browser opens the cache-aside board. Send /users/1, send it again, then open Run.
Press / to search for components. Space pauses, period steps, Backspace deletes the selection, and Ctrl/Cmd+Z undoes that delete. F presents the board. Escape hides panels. Use the Board menu to save, copy a share link, or open a file.
| Template | Explore |
|---|---|
| Cache-aside API | Cache misses, hits, expiry, and eviction |
| Load-balanced API | Routing across backend instances with different capacities |
| Checkout system | Parallel service calls followed by queued fulfillment |
| Redis cache | Commands, memory limits, and eviction policies |
| Kafka stream | Partitions, consumer groups, offsets, commits, and replay |
| Message queue | Acknowledgments, redelivery, and dead letters |
| Primary and replica | Replication delay, read settings, and database capacity |
| API gateway | Rate limits and burst allowances |
| Dependency resilience | Connection failures, timeouts, retries, and circuit breakers |
The Challenges panel includes six incidents, from launch-day overload to checkout under pressure. Each exercise defines a starter system, protected settings, and measurable pass conditions. You can also create an exercise from your own board.
The canvas defines the topology and configuration. A TypeScript engine runs in a Web Worker, advances a virtual clock, and emits execution state. React Flow renders the components and traffic.
flowchart LR
Client --> Backend
Backend --> Cache
Backend --> Database
In the cache-aside template, the backend checks the cache first. A miss triggers a database read and cache fill. Later requests can use the cached record until it expires or is evicted.
Experiments run a captured configuration with a defined workload and optional failure schedule. Reports include response percentiles and count unfinished requests against success. Timings and component behavior come from the simulation model.
| Command | Purpose |
|---|---|
npm run dev |
Start the development server on port 5173 |
npm test |
Run the Vitest suite |
npm run typecheck |
Check TypeScript types |
npm run check |
Run tests and create a production build |
npm run build |
Build and assemble the standalone release |
npm run preview |
Start the standalone production server |
The app uses Next.js, React, and React Flow. Simulation logic and its tests live together under src/simulation/.
src/
app/ Workspace, canvas controls, and browser persistence
board/ Board documents, component rendering, and connections
simulation/ Simulation engine, worker, and component models
challenges/ Exercise definitions, constraints, and verification
interview/ Interview workspace, capacity estimates, and reports
inspector/ Component and connection settings
labs/ Traffic controls and technology labs
scenarios/ Composed example systems
npm run build
npm run previewOpen localhost:3000. Set PORT to use another port.
The build assembles .next/standalone/ with browser assets and public files. To run the release independently, copy that directory and run node server.js inside it.
GitHub Actions runs tests and builds the app. Successful runs upload a liveboard-next-build artifact containing the standalone release. Deployment is separate.


