aquilifer

Aquilifer is a secure LLM provider for websites, the same idea as a crypto wallet extension like MetaMask, but for LLM access instead of a blockchain account. A website never sees your API key. It asks Aquilifer for a completion; Aquilifer holds the credential, enforces the permissions and rate limits you set, and hands back only the response. HOW IT WORKS You connect the LLM provider you actually want to use: an Anthropic API key, an OpenAI key, or a self-hosted/local backend such as llama.cpp, Ollama, or LM Studio. The credential is stored in the extension, never sent anywhere except that provider. A website asks for access through window.aquilifer. The first time, Aquilifer shows a plain-language approval prompt naming the site; you approve or deny it. Once approved, the site can request completions, never the credential itself, and never a way to change which provider answers. WHAT YOU CONTROL - Which provider answers: Anthropic, any OpenAI-compatible endpoint, or a local model, switchable at any time. - Your own usage limits: two-tier rate limits (global and per-interface), configured from the extension, not the website. - A full audit log of every request made on your behalf, viewable from the extension, never fed back into future requests. - Optional password-based encryption for stored API keys. WHAT IT IS NOT Aquilifer requires a pay-per-token API key from the provider you connect. It does not work with a Claude Pro/Max, ChatGPT Plus, or GitHub Copilot subscription; those use a different authentication model. Aquilifer is not affiliated with Anthropic, OpenAI, or any LLM provider. FOR DEVELOPERS No backend to run, no API key to protect server-side, and no per-user cost to carry: the user brings their own credential and pays for their own usage. window.aquilifer exposes a generic chat/stream interface plus provider-specific interfaces (Anthropic Messages, OpenAI Chat Completions) for sites that want a specific provider's native shape. Open source, Apache 2.0 licensed. Full documentation, the page-facing API reference, and an honest list of current limitations are all linked from the extension and at the project website.