AI Coding Agents
Point Claude Code, Codex, Cursor, and other coding agents at the starti.app documentation so they write correct SDK code
AI Coding Agents
If you build with an AI coding agent, it will guess at the starti.app API unless you tell it where the documentation is. Every page on this site is published as plain markdown for exactly that reason.
This guide shows you what to point your agent at, and what to tell it once so you never have to repeat yourself.
Machine-readable documentation
Three endpoints cover every use case:
| URL | What it is | Rough size |
|---|---|---|
/llms.txt | An index of every page — title, description, and a link to its markdown | ~4k tokens |
/llms-full.txt | Every page concatenated into one file | ~92k tokens |
<any page URL>.mdx | The raw markdown of a single page | 1–5k tokens |
The .mdx suffix works on any page. For example, this page is at /sdk/ai-coding-agents.mdx, and the biometrics reference is at /sdk/reference/biometrics.mdx.
Start with /llms.txt and let the agent fetch individual pages on demand. That keeps the context window free for your own code. Reach for /llms-full.txt only when the model has a large context window and you want the whole API available at once.
Tell your agent once
For a single session, this is enough:
Use https://docs.starti.app/llms.txt as the index for the starti.app SDK.
Before writing any SDK code, fetch the .mdx URL of the relevant page and
follow it — do not guess method names or option shapes.Make it permanent
Most agents read a project file on every run. Drop the block below into the right file for your tool and the instruction sticks.
Add it to CLAUDE.md in your project root.
Add it to AGENTS.md in your project root. This is the convention used by OpenAI Codex, Gemini CLI, Kimi, and a growing number of other agents — several of them also read CLAUDE.md as a fallback.
Add it to .cursor/rules/starti-app.mdc in your project root.
## starti.app SDK
This project runs inside a starti.app container — a native iOS/Android app that
wraps our web application and exposes device features to JavaScript.
Documentation:
- Index of every page: https://docs.starti.app/llms.txt
- Any page as raw markdown: append `.mdx` to its URL, e.g.
https://docs.starti.app/sdk/reference/biometrics.mdx
- Read the reference page for a module before calling its methods. Do not
guess method names, parameters, or option shapes.
Rules for SDK code:
- The SDK is loaded from the CDN with a `<script>` tag and exposes a global
`startiapp` object on `window`. Never `import` or `require` it.
- Call `startiapp.initialize()` exactly once, early in the app lifecycle — not
on every route change in a single-page app.
- Native features only exist inside the container. Guard calls with
`startiapp.isRunningInApp()`, or wrap `initialize()` in try/catch, so the
site still works in a normal browser.
- Type definitions come from the `starti.app` npm package. Prefer the types
over assumptions about a signature.Those four rules cover the mistakes agents make most often with this SDK.
Install the type definitions
The single highest-leverage step. The SDK ships full TypeScript types, and an agent with access to them stops inventing method signatures:
npm install --save-dev starti.appThe package is types only — the SDK itself still loads from the CDN at runtime. Once installed, the global startiapp object is typed everywhere, and your agent can read the exact shape of every module from node_modules. See TypeScript Support for details.
Send a single page to a chat
Every page on this site has a Copy Markdown button and a View Options menu at the top. Use them when you want to hand one page to a chat assistant rather than wire up a whole project:
- Copy Markdown — copies the raw markdown of the page to your clipboard
- View as Markdown — opens the
.mdxfile directly - Open in Claude / ChatGPT / Cursor — opens a new conversation prompted to read the current page
Context7
The documentation is also indexed by Context7, which several agents can query through an MCP server. The most complete entry is:
/llmstxt/docs-v2_starti_app_llms_txtContext7 snapshots the docs periodically, so it can lag behind this site. When accuracy matters, fetch the .mdx URL directly.
Keep it current
/llms.txt and every .mdx file are regenerated on each deployment, so they always match what you see on this site. Tell your agent to fetch them rather than rely on what it remembers about starti.app — the SDK gains modules regularly, and a model's training data will be behind.