# WebInfer

> WebInfer is a browser-native protocol for LLM inference. It adds a
> `navigator.llm` API to the page (today via a polyfill; proposed as a W3C
> standard) so a website calls one unified API and the AI runs on
> infrastructure the *user* brings: the WebInfer browser extension configured
> with their own provider keys or local models, Chrome's on-device built-in AI
> (Gemini Nano), or a fallback gateway the *developer* deploys and pays for.
> Keys and prompts stay with the user; the website never ships an API key.
> The SDK mirrors the Vercel AI SDK shape (`generateText`, `streamText`,
> `generateObject`, `embed`, `tool`, …), so existing code ports with
> little change.

The one published package is `webinfer` on npm. Full mechanics — transports,
resolution order, the "works for every visitor" fallback-gateway recipe, tool
calling, on-device AI, and the permission model — are in
[/llms-full.txt](https://webinfer.com/llms-full.txt).

## Start here

- [Quick start](https://webinfer.com/docs/quick-start): 30-second CDN + npm setup and a first `generateText` call.
- [Installation](https://webinfer.com/docs/installation): install the SDK, extension, or daemon.
- [Browser usage](https://webinfer.com/docs/browser-usage): using `navigator.llm` from a web page.
- [API reference](https://webinfer.com/docs/api): every SDK function and its parameters.

## How it works

- [Architecture](https://webinfer.com/docs/architecture): the three-layer model and request flow.
- [Availability](https://webinfer.com/docs/availability): `checkAvailability()`, `promptInstall()`, graceful degradation.
- [Model routing](https://webinfer.com/docs/model-routing): how a request picks a provider/model.
- [Security](https://webinfer.com/docs/security): trust boundaries and the permission model.

## Bring-your-own AI paths

- [Providers](https://webinfer.com/docs/providers): API providers (OpenAI, Anthropic, Groq, …) and local engines (Ollama, LM Studio).
- [Daemon](https://webinfer.com/docs/daemon): the local relay at `localhost:54321`.
- [Gateways](https://webinfer.com/docs/gateways): the fallback-gateway pattern that makes a site work for every visitor.
- [Self-hosting a gateway](https://webinfer.com/docs/gateways/self-hosting): deploy your own gateway in ~15 minutes.

## Capabilities

- [Tools](https://webinfer.com/docs/tools): `tool()` definitions and the client-side agent loop.
- [Vercel AI SDK](https://webinfer.com/docs/vercel-ai-sdk): using WebInfer as an AI SDK provider.

## Full document

- [/llms-full.txt](https://webinfer.com/llms-full.txt): the complete mechanics in one file.