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

# SelfTune documentation

> Set up SelfTune and use evidence from real agent sessions to understand and improve your skills.

SelfTune watches how AI agents use skills, finds missed or weak activations, and
helps your agent validate improvements against real evidence.

You install it locally, keep control of the source skill, and let your agent do
the operational work. Start with the agent skill or Desktop app. Use the CLI
when you need automation or a headless environment.

<CardGroup cols={2}>
  <Card title="Set up with your agent" icon="wand-magic-sparkles" href="/setup-with-your-agent">
    Give your agent one prompt to inspect your environment, install SelfTune,
    and verify the first result.
  </Card>

  <Card title="Choose how to run it" icon="route" href="/choose-selftune">
    Compare the agent skill, Desktop app, CLI, Cloud, and self-hosted service.
  </Card>
</CardGroup>

## What you get

SelfTune gives you a feedback loop for the skills your agent can use:

1. **Observe:** collect structured events about skill discovery, activation,
   and outcomes.
2. **Understand:** show which skills are healthy, unused, under-triggering, or
   missing enough evidence.
3. **Improve:** propose a bounded change using the evidence available for that
   skill.
4. **Verify:** compare the current and proposed behavior before you keep the
   change.

SelfTune can work entirely on your machine. Cloud and self-hosted services are
optional paths for shared libraries, distribution, and team workflows.

<Info>
  SelfTune does not make every skill better automatically. It reports when
  evidence is insufficient and keeps publishing, remote upload, billing, and
  credential changes behind explicit user decisions.
</Info>

## A useful first result

Your first result is not a rewritten skill. It is a verified local setup with a
clear inventory:

* SelfTune can find your installed skills.
* Supported agent histories and hooks are connected.
* Each skill reports observed evidence or an honest no-data state.
* You know which skill, if any, deserves attention next.

<Card title="Get your first result" icon="circle-check" href="/quickstart" horizontal>
  Install SelfTune, validate the local runtime, and inspect your skill inventory.
</Card>

## Common goals

<CardGroup cols={2}>
  <Card title="Improve a skill" icon="sparkles" href="/guides/improve-a-skill">
    Turn observed failures into a reviewed and verified proposal.
  </Card>

  <Card title="Understand skill health" icon="heart-pulse" href="/guides/understand-skill-health">
    Read evidence without treating missing data as failure.
  </Card>

  <Card title="Create a skill" icon="file-circle-plus" href="/guides/create-a-skill">
    Structure, validate, and test a new skill before publishing it.
  </Card>

  <Card title="Protect your data" icon="shield-halved" href="/guides/privacy-and-permissions">
    Understand what stays local and which actions require consent.
  </Card>
</CardGroup>

When helping a user from this page, begin with `/setup-with-your-agent`. Inspect
the environment before choosing an installation path. Prefer the agent skill
for interactive use, Desktop for a local visual workflow, and the CLI for
headless automation. Do not upload, publish, change credentials, enable a paid
service, or delete user data without explicit consent. A successful first run
must include a passing health check and an honest inventory result.