Wollama

⚠️ DISCLAIMER: FOR EDUCATIONAL PURPOSES ONLY This project automates browser interactions with web-based LLM interfaces. This likely violates the Terms of Service of various AI providers (Google, OpenAI, Anthropic, etc.), which prohibit automated access, scraping, and circumventing access controls.

Using this tool may result in:

• Account suspension or permanent ban
• IP blocking
• Legal action from service providers

Do not use this for production, commercial purposes, or any activity that could harm others. If you need programmatic access to these models, use their official APIs:

• Gemini API (generous free tier)
• OpenAI API
• Anthropic API

The authors are not responsible for any consequences resulting from the use of this software.

Turn your browser into a local AI server.

Screencast From 2025-11-21 22-57-19.webm

Wollama (Web + Ollama) is a bridge that exposes web-based LLM interfaces as standard local APIs.

Wollama is architected to be model-agnostic. It acts as a universal adapter layer, allowing you to control any web-based chat interface (like ChatGPT, Gemini, Claude, or Le Chat) using standard tools designed for Ollama.

Mission

To liberate AI models trapped in browser tabs and make them accessible via the command line, IDE plugins, and local scripts.

Current Support

Under the Hood

Wollama uses the Adapter Design Pattern to standardize the chaotic world of DOM manipulation into a clean API.

1. The Server: An HTTP server listening on port 11434.
2. The Interface: It mimics the Ollama API (/api/generate, /api/chat), so existing clients (like ollama run or VS Code extensions) work out of the box.
3. The Backend: Instead of running a heavy model file locally, it sends instructions to a Browser Adapter.
4. The Adapter: Uses Playwright to drive a live Chrome session, typing prompts and scraping responses in real-time.
   • GeminiAdapter: Automates gemini.google.com
   • ChatGPTAdapter: Automates chatgpt.com with proper code block handling

Getting Started

1. Prepare Chrome Start Chrome with remote debugging enabled on port 9222: (this step is optional)

google-chrome --remote-debugging-port=9222

Ensure you are logged into your target platform (gemini.google.com or chatgpt.com).

2. Run Wollama

Start the proxy server:

deno run -A jsr:@sigmasd/wollama
# or
npx rjsr @sigmasd/wollama
# or
bunx rjsr @sigmasd/wollama

3. Usage

Interact with it just like a local Ollama model:

For Gemini:

curl -X POST http://localhost:11434/api/generate -d '{
  "model": "gemini-browser",
  "prompt": "Write a haiku about browser automation.",
  "stream": false
}'

For ChatGPT:

curl -X POST http://localhost:11434/api/generate -d '{
  "model": "chatgpt-browser",
  "prompt": "Explain the adapter pattern in software design.",
  "stream": false
}'

Gemini Chat CLI

Wollama also includes a standalone CLI tool for Gemini that supports file uploads (OCR, PDF analysis, images) and parallel execution.

Usage

Run it directly via JSR:

deno run -A jsr:@sigmasd/wollama/gemini/chat "Explain this document" -f ./doc.pdf

Options

Examples

OCR and Analysis:

deno run -A jsr:@sigmasd/wollama/gemini/chat "OCR this and return text" -f document.pdf

Clean output for piping:

deno run -A jsr:@sigmasd/wollama/gemini/chat -q "Summarize" -f data.csv > summary.md

Parallel execution (Isolated tabs):

deno run -A jsr:@sigmasd/wollama/gemini/chat -n "Task 1" &
deno run -A jsr:@sigmasd/wollama/gemini/chat -n "Task 2" &

Features

• Multi-Model Support: Switch between Gemini and ChatGPT by changing the model name
• Markdown Conversion: Responses are converted to clean markdown with proper code block formatting
• Ollama Compatible: Works with any tool that supports Ollama's API
• Automatic Tab Detection: Finds existing browser tabs or creates new ones

Contributing

We welcome contributions! If you want to add support for Claude or other models, simply implement the Adapter interface and plug it into the main router.

Adding a New Adapter

1. Create a new file: your-model-adapter.ts
2. Implement the adapter interface with ensureReady(), sendMessage(), and close() methods
3. Handle DOM-specific selectors for your target platform
4. Add the adapter to the router in the main server file

Readme created by the gemini-browser model and updated for ChatGPT support.