Inference Bridge

Official reference implementation of the experimental Inference Provider API browser proposal.

Inference Bridge lets websites use AI chat through window.inference — with your permission, on a provider you choose. Bring your own OpenAI, Anthropic, or OpenRouter key, run local Ollama, use on-device browser AI when available, or add an OpenAI-compatible server (LM Studio, llama.cpp, vLLM, and similar). API keys stay in the extension; page scripts never see them. • Streaming chat via window.inference.request() • Experimental function tools via window.inference.experimental (page-executed; keys stay in the extension) • Experimental hosted web search on OpenAI, Anthropic, OpenRouter, and Ollama (Ollama search runs through ollama.com with an optional account key) • Experimental image input and output via window.inference.experimental (vision models / OpenAI & OpenRouter image generation; keys stay in the extension) • Per-site Allow / Deny / Remember prompts • You pick the provider and model • OpenAI, Anthropic, OpenRouter, local Ollama, on-device browser AI (Prompt API), or named OpenAI-compatible endpoints • Optional host access is requested only for each custom server origin you save • Local Ollama and other loopback servers work without special ORIGINS setup • Injects only on https and localhost pages Official reference for the experimental Inference Provider API: https://github.com/SamSamskies/inference-provider-api Privacy policy: https://github.com/SamSamskies/inference-bridge/blob/main/PRIVACY.md