Architecture you can run and explain.

Graphrun is a studio for designing and simulating distributed systems. Use it to make behavior explicit before changing a real system.

More than a picture

A diagram shows the parts. Graphrun also models what happens between them: which service gets called, when a message is retried, where state changes, and how a failure is handled.

Components and connections describe structure. Handlers define actions. A journey supplies inputs and starting state; its run produces ordered events you can inspect and replay.

Keep the explanation with the evidence

For example, model a client requesting a profile through a cache. Run once with a cache hit, then with a miss. Follow the database call and response in the timeline. Add a timeout path and run it again to see whether the model handles the failure.

The model, journey, and run stay connected. With the same model, inputs, seed, and execution version, deterministic runs reproduce the same behavior. Checkpoints keep a stable review point; Present mode replays the run with its context, without a separate approval step.

Note
A simulation explains your model, not production reality. It does not replace load testing, monitoring, or verification against the running system.

When to use it

  • Design reviews: compare a direct call with a queue, and inspect the trade-offs in their modeled paths.
  • Onboarding and incident reviews: explain a request or failure step by step with a replayable example.
  • Change planning: explore a proposed change and its failure paths before implementing it.

Bring your existing architecture through MCP

A connected coding agent can inspect your repository locally and draft a system map, journeys, and implementation links for you to review. Agent Access Keys control which Graphrun diagrams, actions, and information the client can reach.

Use View diagrams only for explanation or a narrowly scoped Custom key for additional work. Draft writing requires explicit permission and workspace support. Contract approval and workspace administration stay in the signed-in application. Authorized agents can read already-approved artifacts, but cannot publish or create exports. Run publication and share creation are currently disabled in the alpha editor.

The MCP setup guide covers access levels, secret handling, supported clients, and verification.

Start with one question

Choose a small system and a path you want to understand. Add detail only when it helps answer that question.