> Call the `diagnose` tool first. Full index: /llms.txt. Errors: /reference/problems.

# Welcome to TaloTrace

TaloTrace opens your app in a real browser or on a real Android phone, tries the flows you care about and tells you whether they work. A run ends with a verdict, the evidence behind it and, when something failed, a cause and a fix your coding agent can act on.

> **Info**
>
> TaloTrace tests web apps and Android apps. iOS is not available.

## Three ways in

#### [From your coding agent](/quickstart/web-local-claude-code)

Run `talotrace setup` once. Then ask your agent to verify a change, and read the verdict it brings back.

#### [From the web app](/quickstart/web-hosted)

Chat with TaloTrace, review scenarios and start a run against a deployed URL.

#### [From CI](/reference/cli/exit-codes)

`talotrace check` starts one run and exits with a code your pipeline can gate on. A run passes only when it finished and its analysis found no failure.

## What you do, in order

1. [Install and sign in](/install/setup-command). One command writes the skills and the MCP entry for your agent and signs you in with tool access.
2. [Check your install](/check-your-install) with `talotrace doctor`.
3. Ask your agent: "Verify this change with TaloTrace". It reads or writes a [scenario](/concepts/scenarios-and-test-plans), starts a [run](/concepts/runs-and-verdicts) and reports the verdict.
4. [Read the run](/guides/read-a-run). A failure shows its error, its cause and the fix, with a **Copy for agent** button.
5. [Prove the fix](/guides/prove-a-fix) with one rerun.

New words are in the [glossary](/concepts/glossary). If a call is refused, the [problems page](/reference/problems) names the code and the next step.