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

# Verifiable Continual Learning for AI Agents

> RELAI turns agent failures and feedback into repeatable learning environments, simulates your agent against them, and optimizes the agent from the results.

<CardGroup cols={3}>
  <Card title="Install & setup" icon="download" href="/installation">
    <span className="card-eyebrow">Start here</span>

    One-line installer, prerequisites, and credentials in three steps.
  </Card>

  <Card title="Quickstart" icon="bolt" href="/quickstart">
    <span className="card-eyebrow">5 minutes</span>

    Run init → simulate → optimize on your own agent project.
  </Card>

  <Card title="All commands" icon="terminal" href="/cli/overview">
    <span className="card-eyebrow">Reference</span>

    Every `relai` command with flags, examples, and failure modes.
  </Card>
</CardGroup>

## The loop, at a glance

Everything in RELAI maps onto a four-step cycle that repeats. You can run it by hand, or drive it from your coding agent with a [plugin](/plugins). For the full picture — an illustration and each component's role — see [The learning loop](/learning-loop).

<Steps>
  <Step title="Initialize">
    From a git-tracked agent project, `relai init` seeds and locally validates a
    simulator harness before registering the agent. Paused runs continue with
    `relai init --resume`.
  </Step>

  <Step title="Author environments & evaluators">
    Capture behavior to learn or preserve as repeatable learning environments — from a prompt, or from a run log plus feedback. Optionally register benchmark suites.
  </Step>

  <Step title="Simulate">
    `relai simulate` measures how the agent behaves today against the environments or benchmarks you select.
  </Step>

  <Step title="Optimize">
    `relai optimize` proposes or applies improvements — and can open a PR with the changes.
  </Step>
</Steps>

<Tip>
  **New here?** No agent of your own yet? [Try the sample agent](/sample-agent) — walk the whole loop in your browser, no install required.
</Tip>