browser-bridge

A local API server + Chrome extension that exposes an OpenAI-compatible API by routing requests through real AI chat interfaces in your browser. No API keys needed for the AI providers themselves — it drives the UI directly.

Your code → POST /v1/chat/completions → FastAPI server → WebSocket → Chrome extension → AI chat tab → response

Works with any tool that speaks the OpenAI API: Python openai SDK, LangChain, LlamaIndex, Continue.dev, Cursor, open-webui, curl, Postman — anything.

---

Supported Models

Provider	Models
Claude (claude.ai)	claude-3-5-sonnet-20241022, claude-3-5-haiku, claude-3-opus, claude-3-sonnet, claude-3-haiku
ChatGPT (chat.openai.com)	gpt-4.1, gpt-4.1-mini, gpt-4.1-nano, gpt-4o, gpt-4o-mini, gpt-4-turbo, o1, o3, o4-mini
Gemini (gemini.google.com)	gemini-2.0-flash, gemini-1.5-pro, gemini-1.5-flash
Copilot (copilot.microsoft.com)	copilot, copilot-gpt-4
Poe (poe.com)	poe-claude-3-5-sonnet, poe-gpt-4o, poe-llama-3, poe-mistral-large
HuggingChat (huggingface.co)	Llama-3.1-70B, Mistral-7B, Phi-3-mini, Command-R+
Le Chat (chat.mistral.ai)	mistral-large, mistral-medium, codestral
You.com	you-gpt-4o, you-claude-3-5
Perplexity	perplexity-pro
Grok (grok.com)	grok-2, grok-2-mini

---

Requirements

• Python 3.10+
• Google Chrome with Developer Mode enabled
• Accounts on whichever AI sites you want to use (free tiers work)

---

Installation

1. Server

cd api-server
pip install -r requirements.txt
python main.py

Server starts at http://localhost:8000. Your API key is printed on first run — it is also visible at the dashboard.

2. Chrome Extension

1. Open chrome://extensions/
2. Enable Developer mode (top right toggle)
3. Click Load unpacked
4. Select the extension/ folder

The extension icon appears in your toolbar. Click it to see connection status and your API key.

3. Log in to AI sites

Open at least one of the supported sites in Chrome and make sure you are logged in. The extension will find and use that tab automatically.

---

Usage

Python (openai SDK)

from openai import OpenAI

client = OpenAI(
    api_key="bb-sk-your-key-here",
    base_url="http://localhost:8000/v1"
)

# Chat completions
response = client.chat.completions.create(
    model="gpt-4.1-mini",
    messages=[{"role": "user", "content": "Hello"}]
)
print(response.choices[0].message.content)

# Responses API (also supported)
response = client.responses.create(
    model="gpt-4.1-mini",
    input="Hello"
)
print(response.output[0].content[0].text)

Interactive test client

# Single message
python test_client.py --prompt "What is 2+2?"

# Interactive chat with persistent memory
python test_client.py --chat --model gpt-4.1-mini

# Use a specific model
python test_client.py --chat --model claude-3-5-sonnet-20241022

# With a system prompt
python test_client.py --chat --system "You are a cybersecurity expert"

# Resume a previous session
python test_client.py --chat --session my-session

# List all saved sessions
python test_client.py --sessions

# Clear a session
python test_client.py --clear --session my-session

PowerShell

$res = Invoke-RestMethod `
  -Uri "http://localhost:8000/v1/chat/completions" `
  -Method POST `
  -Headers @{ "Authorization" = "Bearer bb-sk-your-key-here" } `
  -ContentType "application/json" `
  -Body '{"model":"gpt-4.1-mini","messages":[{"role":"user","content":"Hello"}]}'

$res.choices[0].message.content

curl

curl http://localhost:8000/v1/chat/completions \
  -H "Authorization: Bearer bb-sk-your-key-here" \
  -H "Content-Type: application/json" \
  -d '{"model":"gpt-4.1-mini","messages":[{"role":"user","content":"Hello"}]}'

---

Dashboard

Open http://localhost:8000/dashboard in any browser while the server is running.

Tab	What it does
Overview	Live connection status, active key, supported sites
API Keys	Generate single or bulk keys in 10 different styles
Models	Browse all 45+ supported models
Test API	Fire a live request directly from the browser

---

API Keys

Keys are fake — they exist only for local auth and tool compatibility. Generate keys in any style to match whatever tool you are integrating with:

# Single key
curl -X POST http://localhost:8000/v1/keys \
  -H "Content-Type: application/json" \
  -d '{"style":"openai","label":"my-tool"}'

# Bulk — multiple styles at once
curl -X POST http://localhost:8000/v1/keys/bulk \
  -H "Content-Type: application/json" \
  -d '{"keys":[{"style":"openai","count":3},{"style":"anthropic","count":2}]}'

Supported styles: browser-bridge, openai, anthropic, google, huggingface, mistral, cohere, copilot, perplexity, grok

---

How Tab Management Works

The extension uses sessionStorage fingerprinting to track every tab it opens:

1. A new request comes in for a model (e.g. gpt-4o)
2. Background script checks its registry for an existing tab on chat.openai.com
3. If a tab exists, it asks the content script: "Is this tab ours, and has it been used?"
   • Ours + unused → reuse the tab as-is, inject prompt directly
   • Ours + used → click "New Chat" inside the same tab, then inject
   • Not ours → open a brand new tab (won't interfere with your own browsing)
4. If no tab exists → open a new tab, stamp it, inject prompt

This means you can keep browsing normally — the tool only ever uses tabs it opened itself.

---

Pipeline & Filtering

pipeline.py provides input/output filtering and persistent session context for the test client:

• Input filter — strips null bytes, normalises whitespace
• Output filter — removes scraped UI artifacts (buttons, labels, navigation text that sometimes leaks into DOM scraping)
• Session context — per-session chat history stored in session_context.json, survives restarts

---

Endpoints

Method	Path	Description
GET	/	Health check
GET	/dashboard	Web dashboard
GET	/v1/status	JSON status (extension connected, keys, models)
GET	/v1/models	List all models
POST	/v1/chat/completions	OpenAI chat completions
POST	/v1/completions	Legacy text completions
POST	/v1/responses	OpenAI Responses API
POST	/v1/embeddings	Stub embeddings (zero vectors)
POST	/v1/keys	Generate a key
POST	/v1/keys/bulk	Generate multiple keys
GET	/v1/keys	List all keys
DELETE	/v1/keys/{key}	Revoke a key
WS	/ws/extension	Extension WebSocket

---

Project Structure

browser-api-bridge/
├── api-server/
│   ├── main.py              # FastAPI server, all endpoints
│   ├── websocket_manager.py # WebSocket connection to extension
│   ├── queue_manager.py     # Serial request queue
│   ├── keygen.py            # Fake API key generator
│   ├── pipeline.py          # Input/output filters + session context
│   ├── test_client.py       # CLI test client
│   ├── dashboard.html       # Web dashboard (served at /dashboard)
│   └── requirements.txt
└── extension/
    ├── manifest.json        # MV3 manifest
    ├── background.js        # Service worker, WebSocket, tab registry
    ├── content.js           # DOM automation, fingerprinting
    ├── popup.html           # Extension popup
    └── popup.js

---

Limitations

• Responses depend on the AI site's UI — if a site updates its DOM structure, selectors may need updating
• One request at a time (serial queue) — parallel requests are queued and processed in order
• Requires the browser to stay open while the server is running
• Free tier rate limits of the AI sites still apply

---

License

MIT