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

# AI Track Guidance

Use this page if you are building for the AI track and want a practical model
for how AI should fit into your Initia appchain architecture.

You do not need to run model inference onchain. In most projects, inference
will happen offchain, with the appchain handling the parts of the product that
need blockchain guarantees.

<CardGroup cols={2}>
  <Card title="Use AI Offchain" icon="robot">
    Generate or transform content, make recommendations, power copilots or
    agents, and run classification or summarization.
  </Card>

  <Card title="Use the Appchain" icon="diagram-project">
    Store ownership and state, handle rewards and payments, enforce access or
    reputation, and coordinate marketplaces or escrow.
  </Card>
</CardGroup>

## External AI Services

You may use hosted model APIs in your project. If your project depends on a
third-party AI provider, plan to supply your own API key or backend
configuration for development and demos. For most hackathon teams, hosted APIs
are the fastest and most common way to add AI features.

<Warning>
  **API key security:** If your project uses an external AI API, store secrets
  in a local `.env` file or another secure secret manager. Do not commit API
  keys to GitHub or expose them in your demo video, screenshots, or frontend
  code.
</Warning>

If you need to add a model provider to your app, start with the official API
docs for your chosen service:

<CardGroup cols={3}>
  <Card title="OpenAI" icon="book-open" href="https://platform.openai.com/docs/quickstart">
    API docs
  </Card>

  <Card title="Anthropic" icon="book-open" href="https://platform.claude.com/docs/en/api/overview">
    API docs
  </Card>

  <Card title="Google Gemini" icon="book-open" href="https://ai.google.dev/docs">
    API docs
  </Card>
</CardGroup>

<Note>
  For demos, it is acceptable to use mocked, cached, or pre-generated AI
  outputs if you clearly disclose that setup. Judges will evaluate the
  product, appchain integration, and overall user experience, not whether the
  model itself is hosted onchain.
</Note>