Analysis updated 2026-05-18
Automate repetitive browser tasks like deleting spam or filling forms by typing a plain-language command.
Use a local Ollama model to browse with AI assistance, keeping all data on your own machine without any cloud account.
Fetch a YouTube video transcript and summarize it without leaving the browser.
Compare prices or gather information across multiple websites by describing what you want in plain English.
| alexvilelabah/bah-browser | rosekhlifa/image-studio | ellian-eorwyn/hephaestus | |
|---|---|---|---|
| Stars | 51 | 50 | 53 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | easy | easy |
| Complexity | 2/5 | 3/5 | 2/5 |
| Audience | vibe coder | developer | vibe coder |
Figures from each repo's GitHub metadata at analysis time.
Windows only, the unsigned installer triggers a SmartScreen warning that requires clicking 'More info' then 'Run anyway'.
Bah is a desktop web browser that lets an AI agent operate the web on your behalf. Instead of clicking and typing yourself, you type a plain-language command in a side panel and the AI reads the page, identifies interactive elements, clicks, types, and navigates until your request is done. It is inspired by Perplexity Comet and is open source, built for Windows and distributed as a one-click installer. The browser ships free out of the box. For basic chat and image generation, no account or API key is required. If you want the full autonomous agent or faster responses, you can connect a cheap API key from DeepSeek (the recommended option), Mistral, or NVIDIA NIM. If you prefer to keep everything on your own computer without sending data anywhere, Bah can download and run an Ollama model locally, entirely offline. Each browser tab has its own conversation with the AI. The agent works by observing the page, taking a screenshot, deciding on one action at a time, and executing it using real operating system input rather than synthetic events, so modern frameworks and anti-bot protections treat it like a real user. It supports up to 25 steps per task and uses a structured action vocabulary: click by element ID, fill a form field, click by visible text, go to a URL, open or close tabs, and signal when it is finished. Beyond the autonomous agent, Bah includes practical built-in shortcuts that cost no AI tokens: opening many YouTube videos at once, jumping to a specific spoken phrase in a video, or fetching a transcript for a quick summary. It also ships with a full ad blocker and a daily-updated list of malicious URLs for safer browsing. The tech stack is Electron with Chromium at the shell layer, React 19 and TypeScript for the UI, and Vite as the build tool. The app currently runs on Windows and auto-updates silently in the background. Source code is on GitHub.
Bah is a desktop AI browser where you type plain-language commands and an AI agent clicks, types, and browses the web for you, it works free out of the box or with a cheap API key.
Mainly TypeScript. The stack also includes TypeScript, Electron, React.
Source code is publicly available but the specific license terms are not clearly stated in the README.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.