stateweave Docs
Get started

StateWeave documentation

Test your AI agent on working copies of the apps it uses, filled with a realistic company's data. Your agent runs where it always runs. StateWeave hands each run fresh apps, records every call, and grades what the agent left behind.

QuickstartInstall sw, sign in and run your agent on its first test in five minutes. Connect your agentWhat a run hands your agent, and how it hands an answer back. Works with any language. Apps and credentialsPoint the official Slack, Google, Salesforce, Shopify and Zendesk clients at a run. API referenceEvery sw command, the Python SDK and the REST API.

How it works

A scenario is a company that never existed: its people, customers, history and the data each of its apps holds. A test is a request someone could give your agent in that company, with an answer key. A run is one attempt by your agent at one test.

1Start a runStateWeave copies the test's apps from the scenario, so every run starts from the same state.
2Your agent worksIt calls the apps' real APIs or MCP servers with the credentials the run hands over.
3Hand in the answerFor a question, the answer. For a change, nothing: the apps hold the result.
4Graded and shownThe run passes or fails on the apps' end state and the answer, with every call on its page.

Your agent never runs on StateWeave. It stays on your laptop, in CI or wherever you deploy it, and reaches the apps over HTTPS. That means you test the real thing: your prompts, your tools, your model and your harness.

Apps

Each app is a twin of the real service: the same endpoints, request and response shapes, errors and limits, so the vendors' own client libraries work against it unchanged.

AppAPI your agent callsMCP server
SlackWeb APIYes
GmailGmail API v1Yes
Google CalendarCalendar API v3No
Google SheetsSheets API v4, with Drive API v3 for filesNo
Google DriveDrive API v3No
SalesforceREST API v68.0Yes
ShopifyAdmin GraphQL API 2026-10Yes
ZendeskSupport API v2No

See Apps and credentials for each app's addresses and client setup, and MCP servers for agents that use tools over MCP.

Three ways to connect

Pick whichever fits your harness. They all start the same runs and show up on the same pages.

  • sw run runs any command as the agent and passes the run to it in environment variables. Nothing to change in your code but where it reads its task. Learn more.
  • The Python SDK gives you the run as an object inside your own harness: run.prompt, run.apps, run.finish(answer). Learn more.
  • The REST API is four calls: list tests, start a run, finish it, read the result. Learn more.

Get access

StateWeave is in early access. Create an account and request a workspace. Once it is approved, your workspace has its own address, such as https://acme.app.stateweave.net, where you build scenarios, write tests and watch runs. Questions: mikewei@stateweave.net.