AirmailAI

AirmailAI is an open-source, bring-your-own-key chat client for using several AI providers from one browser interface. It supports OpenAI, Anthropic, Google, and OpenRouter, so a user can move among their available models without maintaining a separate chat application for each provider. There is no AirmailAI account and no AirmailAI subscription. Users supply API credentials from the providers they want, then pay those providers for the tokens they consume. The product is aimed at people who value a lightweight common interface and want their conversations kept out of another hosted application database.

The product has a two-part architecture. A static website supplies the chat interface, model choices, provider configuration, and other visible controls. A Chromium browser extension acts as the local backend. It stores API keys and conversation data, accepts requests from the official AirmailAI site, and sends each request directly to the chosen provider over HTTPS. The provider streams its response back through the extension to the page. AirmailAI says its own servers never receive the messages, attachments, system prompts, model settings, or raw API keys involved in a conversation.

The extension is also the boundary that keeps saved credentials away from normal website code. Its protocol lets the AirmailAI page save a key, check whether one exists, ask the extension to make a provider request, or delete the key. It does not include a method for returning the stored raw key to the website. Keys remain in non-synchronized extension storage, while chats, attachments, and file metadata stay in a local IndexedDB database. Interface settings may sync through a signed-in browser profile, but chat history, files, and API credentials do not. Users can export individual chats or perform a bulk export, import AirmailAI, LM Studio, and SillyTavern histories, and delete local data from the settings.

As a chat client, AirmailAI goes beyond plain text prompts. The site lists file uploads and web search, with actual file support depending on the selected provider. OpenRouter can provide access to additional models, including some free options, while the other supported APIs use their own billing and feature rules. AirmailAI does not host models, run inference, or control the answers it displays. That keeps the client focused, but it also means provider terms, retention settings, supported file types, tool behavior, and model quality still govern anything sent outside the browser.

The security design addresses content as well as credentials. Release builds verify the identity of both sides of the website and extension connection, and the extension accepts external messages only from the official site. Imported chats are parsed into a limited schema with bounds on files, chats, messages, and field sizes. Markdown messages are sanitized before display, while raw HTML is shown as text instead of being executed. AirmailAI also says the application does not load third-party analytics or remotely hosted application scripts. These measures reduce exposure inside the client, though they cannot protect data from someone who controls the device or browser profile. Once a request reaches a provider, that provider's security and data policies take over.

The current experience is designed for desktop Chromium-based browsers such as Chrome, Edge, and Brave, and the extension is required. It does not currently work on mobile. That is a meaningful constraint for anyone who expects the same chat history across a phone, laptop, and multiple browser profiles. Local storage also makes the user responsible for exports and device continuity. In return, the setup avoids creating another account, installing a full desktop application, or routing conversations through an AirmailAI backend. It is a particularly good fit for technical users who already understand provider API keys and prefer usage billing to several overlapping chat subscriptions.

AirmailAI itself is free and published under the Apache 2.0 license. It does not add a markup or per-request fee, although paid model usage is still billed directly by the relevant provider and consumer chat subscriptions do not include API access. Users who want to avoid paid inference can choose free models available through OpenRouter, subject to those providers' data practices. The project accepts bug reports and feature requests through GitHub. Its security documentation is unusually specific about browser boundaries, local storage, untrusted chat content, and the point at which responsibility passes to the model provider.