Bah
An open-source AI browser that operates the web from plain-language commands, reading the page and clicking, typing and scrolling with a real mouse and keyboard, using cloud models such as DeepSeek or a local Ollama model.
Bah is an Electron-based browser with an agent panel: a user types a command, such as opening an email client and deleting spam, and the AI decides step by step what to do until the task is finished, reading the page's structure, numbered interactive elements and on-screen text through OCR rather than only looking at a screenshot, then acting through structured tools.
Each browser tab keeps its own AI conversation, remembered separately from other tabs, and one-shot skills include opening several videos at once, jumping to the moment a phrase is spoken across multiple videos, and chatting about a YouTube video's content. Image generation works with no key needed; chat and the full autonomous agent need a low-cost key from a supported cloud provider such as DeepSeek, Mistral or NVIDIA NIM, or can run entirely offline and free through a local Ollama model pulled directly from the app's built-in AI panel.
It ships as a Windows installer with automatic update checks, and its source is also available to build and run directly; it is written in TypeScript and React.