Workbench
Where to do the work
Once you have a program, you need somewhere to edit it, query it, and check it, on your own or with an agent. Each place below runs the same compiled program and the same checks.
- WebMCP workbench: an API program with every check, for you and a connected agent.
- Data app editor: edit a data app’s two files and run it with Why.
- Explorer playground: query a program’s structure.
- Lean proofs in the browser: check a supported law with Lean.
Connect your agent once, then any of these can be driven from it. Inspect, edit, and recheck walks one full session; the worked tool invocations show each request.
Questions the checks answer
Ask your agent the question in your own words. It runs the matching check and tells you what the answer covers.
- Does this API satisfy its applicable checks?
- Do routes declare the operations they perform?
- Do database operations respect the resource policy?
- Did I compose IO incorrectly?
- Is causal order consistent? Can modeled operations race?
- Can failures and message ordering break my distributed model? Run the distributed-systems examples
- Do transaction retries and uncertain commits preserve my invariants?
- Which operations fit within my resource bounds? Run the ledger admission demo
- What properties should I test?
- Does my function satisfy a declared law?
- Has Lean checked a supported law?
- Does the handler follow my model of state changes?
- Do my discharged obligations suggest a faster commit strategy?
- Do related mutations follow the structure I declared? Run the flow tutorial
- Does an agent’s report agree with computed program facts? Run the claim-checking example
- Can I inspect a declared system-property verdict?
Explore a running application
- Notes: an owner-scoped collection with list, write, add, and clear routes.
- ProjectHub: a service with several entities and a generated frontend.
- Program reports: routes, tables, guarantees, and the evidence behind them.
Ask your agent about anything here; it has the detail.