> ## Documentation Index
> Fetch the complete documentation index at: https://runmirrors.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# What is Mirrors?

> Catch broken agents before your customers do.

Mirrors replays your AI agent's real production sessions against your next change — a new prompt, tools, or model — and shows you exactly what would have broken. Nothing real gets touched.

Your agent takes real actions on real systems, so testing changes against live production is slow, risky, and irreproducible. Mirrors ingests your agent's traces and builds a runnable copy of its world: a seeded database plus bound tools. Agents replay against the mirror, never your real systems, so even destructive flows (refunds, deletes, sends) are safe to test.

## How it works

<Steps>
  <Step title="Add two lines">
    The [mirrorkit collector](/collectors/python) streams your agent's traces to Mirrors. It works with traces from your agent framework or observability platform too.
  </Step>

  <Step title="Mirrors builds your agent's world">
    An isolated, runnable copy of the environment: a seeded database plus bound tools, each scored for [fidelity](/how-it-works#fidelity-scoring) against the real traces.
  </Step>

  <Step title="Replay any session against any change">
    From the dashboard, [CLI](/cli), [MCP client](/mcp-server), or [CI](/ci-gate). Same session, two versions of the agent, one caught regression. Production is never touched.
  </Step>
</Steps>

## Start here

<CardGroup cols={2}>
  <Card title="Quickstart" icon="rocket" href="/quickstart">
    Connect your agent and gate merges in CI in a few minutes.
  </Card>

  <Card title="How it works" icon="lightbulb" href="/how-it-works">
    Mirrors, fidelity scoring, deterministic seeding, and evals.
  </Card>

  <Card title="MCP server" icon="plug" href="/mcp-server">
    Drive Mirrors from Claude Code, Cursor, VS Code, ChatGPT, and more.
  </Card>

  <Card title="API reference" icon="code" href="/api-reference/introduction">
    The versioned /v1 API, authenticated by workspace keys.
  </Card>
</CardGroup>

## What you can do with it

* **Catch regressions before they ship** — replay real sessions against every change and see exactly what would have broken.
* **Reproduce the bug that paged you**, on demand and byte-for-byte, via deterministic seeding.
* **Test risky and destructive flows** (refunds, deletes, sends) safely against the mirror instead of live systems.
* **Gate CI on replays** so every change is checked before it merges.
* **Measure whether an agent change is actually better** with per-tool coverage and accuracy scores.
* **Drive mirrors from your own code** over a versioned [/v1 API](/api-reference/introduction) authenticated by workspace keys.

<Note>
  Privacy by construction: the mirror keeps only the skeleton of which entities exist and how they connect. Every actual value is fabricated, so redacted sessions work fine.
</Note>