OpenBrowser

Turn free browser AI chat into a local coding agent — no API key, no extra AI subscription.

OpenBrowser is a free AI tool and local AI agent harness that connects ChatGPT, Gemini, DeepSeek, Claude, Perplexity, GLM, Grok, and other browser-based AI assistants to your project workspace. Run commands in the terminal — prompts are auto-sent to your AI tab via the browser extension, responses flow back to the terminal automatically. No manual copy-paste. No API keys required — use your existing browser AI subscription (including free plans).

Star the repo on GitHub to support the project and help others discover it.

What is OpenBrowser?

OpenBrowser is an open-source AI agent for developers who want ChatGPT for free coding, Claude for free coding, or any browser AI plan — without paying for a separate API key or AI subscription. It acts as a CLI harness between your terminal and the AI chat tab you already use in Chrome.

Free AI tool — works with free browser tiers (ChatGPT, Claude, Gemini, and more)
No API key — your browser session is the connection; no OpenAI or Anthropic API billing
AI coding agent — ask questions, edit files with diff preview, attach @file / @folder context
Local-first — bridge runs on 127.0.0.1; you approve every file change

FAQ

Can ChatGPT be used for coding?

Yes. ChatGPT is widely used for coding help, but the browser UI is not built for editing a whole project. OpenBrowser lets you use ChatGPT for coding from the terminal: send prompts from your repo, get answers in the terminal, and in agent mode apply multi-file changes with diff preview — still using your ChatGPT browser subscription, not an API key.

How to use ChatGPT for free coding?

Open ChatGPT in Chrome (free or Plus plan).
Install the OpenBrowser Chrome extension and run openbrowser in your project folder.
Run openbrowser ask "..." for Q&A or openbrowser agent "..." for file edits.

You get a free AI coding workflow as long as your browser plan allows chat — no separate API subscription.

How to use Claude for free coding?

Same flow with Claude: keep a Claude tab open, load the extension, run openbrowser ask or openbrowser agent from your project. OpenBrowser auto-sends prompts and returns responses to the terminal so you can use Claude for free coding (or on your existing Claude subscription) without an Anthropic API key.

Do I need an API key or paid AI subscription?

No API key is required. OpenBrowser does not call OpenAI, Anthropic, or Google APIs directly — it bridges your existing browser AI subscription. If you already use ChatGPT, Claude, or Gemini in the browser (including free tiers), you can use them as your AI agent for local development.

Is OpenBrowser a free alternative to API-based coding agents?

OpenBrowser is a free, open-source AI tool for developers who prefer browser AI over API billing. It is not a replacement for every feature of paid IDE agents, but it delivers ask mode, agent mode with diffs, and project context — powered by the AI subscription you already have in the browser.

What is an AI harness?

In this project, an AI harness is the local bridge (CLI + extension) that wires browser chat to your workspace: queue prompts, inject them into the AI composer, capture replies, and apply file operations safely on your machine.

Features

Mode	What it does
Ask	Auto-send prompt with system instructions; Markdown response appears in the terminal
Agent	Auto-send task + project context + JSON schema instructions; preview diffs, apply or reject
Server	Run the bridge API on `http://127.0.0.1:5000` for the browser extension

Local bridge server (Fastify) on port 5000
Interactive wake mode — run openbrowser with no args for an ask/agent menu
@file / @folder context attachments with Tab completion
Large prompts auto-delivered as openbrowser-prompt.txt when they exceed the UI paste limit
Zod-validated AI operation schema (CREATE_FILE, EDIT_FILE, etc.)
Unified diff preview before any file is touched
Edit history stored in .openbrowser/history.json
Chrome extension for supported AI sites

Requirements

Node.js 20 or later
pnpm 11.x (install guide)
Google Chrome (or any Chromium browser) for the extension

Quick Start

# 1. Clone and install
git clone https://github.com/1129Aliasgar/OpenBrowser.git
cd OpenBrowser
pnpm install

# 2. Build the CLI
pnpm build

# 3. Enable the openbrowser command in your terminal
pnpm setup
pnpm link --global

# 4. Copy environment config
cp .env.example .env        # Windows: copy .env.example .env

# 5. Load the Chrome extension (see Browser Extension section)

# 6. Open https://chatgpt.com in Chrome and reload the tab

# 7. Open a new PowerShell window, then run
openbrowser ask "How do I add JWT auth in Express?"

Before running: An AI chat tab must be open in Chrome with the extension loaded. The extension listens over SSE and auto-injects prompts.

Getting openbrowser is not recognized?
Run pnpm setup, then pnpm build and pnpm link --global, and open a new terminal. Or use pnpm start ask "..." without global install — see CLI Usage.

Install pnpm

corepack enable
corepack prepare pnpm@11.0.0 --activate

Verify:

node -v    # v20+
pnpm -v    # 11.x

Installation

1. Install dependencies

pnpm install

2. Configure environment

cp .env.example .env        # Windows: copy .env.example .env

Variable	Default	Description
`PORT`	`5000`	Bridge server port
`BRIDGE_TOKEN`	(optional)	Bearer token for `/operations` and `/session/prompt` requests
`PROMPT_INJECTION_CHAR_LIMIT`	`12000`	Above this length, prompts are sent as a `.txt` file attachment

3. Build

pnpm build

Compiled output is written to dist/.

CLI Usage

Make `openbrowser` available in your terminal

The openbrowser command is not installed automatically. PowerShell shows:

openbrowser: The term 'openbrowser' is not recognized ...

That is expected until you complete the setup below.

Option A — Global command (recommended)

One-time pnpm PATH setup (required on Windows):

pnpm setup

Close and reopen PowerShell after this command. It adds %LOCALAPPDATA%\pnpm to your PATH.

Register the CLI (run from the project root after every fresh clone):

pnpm build
pnpm link --global

Verify in a new terminal:

openbrowser --help

Option B — Run without global install

From the project root:

# Compiled CLI (requires pnpm build first)
pnpm start ask "Explain this repo structure"
pnpm start agent "Add input validation"
pnpm start server

# Dev mode with auto-reload (no build needed)
pnpm dev
pnpm exec tsx src/index.ts ask "Explain this repo structure"

Option C — Interactive dev mode

pnpm dev

Starts the CLI in watch mode. With no arguments, you get an interactive menu:

Select mode:
  1. ask
  2. agent
  q. exit
mode>

Use @path/to/file in the prompt area for context attachments. Tab completes paths after @.

Commands

openbrowser                          # Interactive mode (ask / agent menu)
openbrowser ask "<prompt>"           # Ask mode — Q&A in terminal
openbrowser agent "<task>"           # Agent mode — file ops with diff preview
openbrowser server                   # Run bridge server only (no CLI workflow)
openbrowser --help                   # Show all commands
openbrowser --version                # Show version

Ask mode

openbrowser ask "How do I implement rate limiting in Fastify?"

Bridge server starts on port 5000.
CLI queues the prompt with a system instruction (Markdown answer expected).
Chrome extension receives the job instantly via SSE, injects the message into the AI composer, and clicks Send.
When the AI finishes replying, the extension captures the text and posts it to the bridge.
CLI receives the response over SSE and prints it in the terminal.

Agent mode

openbrowser agent "Add a health check endpoint to the server"

Project context is generated from the current directory.
CLI queues task + context + JSON schema system instructions.
Extension auto-sends to the AI tab and waits for a JSON response with conversationId.
CLI validates the JSON, shows unified diffs for each proposed change.
Confirm with y to apply, or N to reject.

All applied changes are logged under .openbrowser/history.json.

Long prompts (file attachment)

When a prompt (including system instructions and @ context) exceeds 12,000 characters by default, OpenBrowser saves the full text to .openbrowser/prompts/<session>.txt and the extension uploads it as openbrowser-prompt.txt instead of pasting into the composer. A short note is sent in the text field telling the AI to read the attachment.

Override the limit with PROMPT_INJECTION_CHAR_LIMIT in .env.

Bridge Server

The bridge server is the local API that connects the CLI, browser extension, and file operations.

Start the server

Dedicated server command (keeps running until you stop it):

openbrowser server

Dev watch mode (auto-restarts on file changes):

pnpm dev:server

Embedded server — openbrowser ask and openbrowser agent start the server automatically and shut it down when the command finishes.

Default URL

http://127.0.0.1:5000

Endpoints

Method	Path	Description
`GET`	`/health`	Health check (used by the extension popup)
`GET`	`/summary`	Project context summary
`POST`	`/session/prompt`	CLI submits prompt job with system instructions
`GET`	`/session/:id/events`	CLI SSE stream for completed response
`GET`	`/session/:id/status`	Poll session status
`GET`	`/browser/events`	Extension SSE stream for new prompt jobs
`POST`	`/browser/claim`	Extension claims a job before processing
`GET`	`/browser/prompt-file/:id`	Extension downloads prompt `.txt` for attachment
`POST`	`/browser/chunk`	Extension streams partial ask-mode responses
`POST`	`/browser/response`	Extension posts AI reply back to bridge
`POST`	`/operations/preview`	Preview diffs for operations
`POST`	`/operations/apply`	Apply validated operations

Verify the server is running

curl http://127.0.0.1:5000/health

Expected response:

{ "status": "ok" }

Or open http://127.0.0.1:5000/health in your browser.

Browser Extension (Chrome)

The extension watches supported AI chat pages, injects prompts, captures responses, and forwards them to the local bridge server.

Supported sites

Provider	URL
ChatGPT	chatgpt.com
Gemini	gemini.google.com
DeepSeek	chat.deepseek.com
Claude	claude.ai
Perplexity	perplexity.ai
GLM	chat.z.ai
Grok	grok.com

Load the extension in Chrome

Start the bridge server (extension needs port 5000):
```
openbrowser server
```
Open Chrome extensions page:
```
chrome://extensions
```
Enable Developer mode (toggle in the top-right corner).
Click "Load unpacked".
Select the browser-extension folder inside this repo (the folder that contains manifest.json).
Pin the extension — click the puzzle icon in the Chrome toolbar → pin OpenBrowser Bridge.
Verify connection — click the extension icon. The popup should show:
```
Bridge server is running on port 5000.
```
If it says "Bridge server is not reachable", make sure openbrowser server is running.

How the extension works

You run openbrowser ask or openbrowser agent in your project directory.
The CLI submits a prompt job to POST /session/prompt on the bridge server.
The bridge pushes the job instantly over SSE to GET /browser/events.
The content script claims the job, injects the message (or attaches openbrowser-prompt.txt for long prompts), and clicks Send.
When the AI reply is complete, the extension posts it to POST /browser/response.
The CLI receives the result over GET /session/:id/events and prints it in the terminal.

Important: Keep an AI chat tab open and reload it after updating the extension.

AI Response Format

Agent mode expects JSON like this:

{
  "operations": [
    {
      "action": "CREATE_FILE",
      "path": "src/example.ts",
      "content": "export const hello = 'world';\n"
    }
  ],
  "conversationId": "550e8400-e29b-41d4-a716-446655440000"
}

Supported actions: CREATE_FILE, EDIT_FILE, DELETE_FILE, RENAME_FILE, CREATE_FOLDER.

Paths must be relative to the project root. Directory traversal (../) is rejected.

Development

pnpm dev           # CLI in watch mode (tsx)
pnpm dev:server    # Bridge server in watch mode
pnpm build         # Compile TypeScript → dist/
pnpm typecheck     # Type-check without emitting
pnpm test          # Run Vitest unit tests
pnpm test:watch    # Vitest in watch mode

Project layout

OpenBrowser/
├── src/
│   ├── assest/           # Logo, banner, favicon
│   ├── index.ts          # CLI entry point
│   ├── server/           # Bridge server (Fastify + SSE)
│   ├── context/          # Project context & @ attachments
│   ├── protocol/         # Zod schemas & validation
│   ├── parser/           # AI response parsing
│   ├── operations/       # Diff preview & file executor
│   ├── memory/           # .openbrowser storage
│   └── shared/           # Terminal UI, prompt delivery
├── browser-extension/    # Chrome MV3 extension
├── dist/                 # Compiled output (after pnpm build)
├── .env.example
├── CONTRIBUTING.md
├── CODE_OF_CONDUCT.md
├── SECURITY.md
├── LICENSE
├── package.json
└── pid.md                # Full product specification

Troubleshooting

`openbrowser` is not recognized (PowerShell)

Cause	Fix
pnpm global bin not in PATH	Run `pnpm setup`, restart PowerShell
Project not built	Run `pnpm build`
CLI not linked globally	Run `pnpm link --global`, then open a new terminal
Want to skip global install	Use `pnpm start ask "..."` or `pnpm dev`

Timed out waiting for browser AI response

Open a supported AI site in Chrome (not Edge/Firefox alone).
Reload the AI tab after installing the extension.
Reload the extension on chrome://extensions.
Confirm the popup shows "Bridge running" and at least one AI tab ready.
Run openbrowser from the project root — ask/agent start their own bridge server automatically.

Extension shows "Bridge server is not reachable"

Confirm the server is running (started automatically by openbrowser ask / openbrowser agent).
Check port 5000 is free: curl http://127.0.0.1:5000/health
Reload the extension on chrome://extensions
Leave BRIDGE_TOKEN unset for local dev (extension does not send auth headers)

Long prompt not attaching

Confirm the provider supports file upload (ChatGPT, Claude, Gemini, DeepSeek).
Check .openbrowser/prompts/ for the saved session file.
Lower PROMPT_INJECTION_CHAR_LIMIT to test, or shorten @ context attachments.

Agent mode rejects the AI response

Response must be valid JSON with operations and conversationId (UUID v4).
No markdown code fences unless the JSON is inside them.
Every path must stay inside the project root.

Changes not applied

You must confirm with y after reviewing diffs.
Check .openbrowser/history.json for the operation log.
Run agent mode from the project root you want to modify.

Architecture

Terminal (CLI)  ──POST /session/prompt──►  Bridge :5000
       ▲                                        │
       │ SSE /session/:id/events                │ SSE /browser/events
       │                                        ▼
       └────── POST /browser/response ◄──  Chrome Extension
                                                  │
                                                  ▼
                              AI chat tab (ChatGPT, Gemini, Claude, …)

CLI — submits prompts with system instructions, waits for responses
Bridge Server — session queue, job dispatch, prompt file storage, response delivery
Browser Extension — SSE job delivery, composer injection / file attach, response capture
Context Engine — reads workspace and builds agent prompts
Operation Executor — applies approved file changes

For the full product specification, see pid.md.

Contributing

OpenBrowser is open source and welcomes contributions.

Read CONTRIBUTING.md for setup and PR guidelines.
Follow the Code of Conduct.
Report security issues privately — see SECURITY.md.

Quick start for contributors:

git clone https://github.com/1129Aliasgar/OpenBrowser.git
cd OpenBrowser
pnpm install && pnpm build && pnpm test

Fork the repo, create a branch, and open a pull request on GitHub.

Name		Name	Last commit message	Last commit date
Latest commit History 27 Commits
.github/ISSUE_TEMPLATE		.github/ISSUE_TEMPLATE
browser-extension		browser-extension
src		src
.env.example		.env.example
.gitignore		.gitignore
.npmrc		.npmrc
CODE_OF_CONDUCT.md		CODE_OF_CONDUCT.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
package.json		package.json
pid.md		pid.md
pnpm-lock.yaml		pnpm-lock.yaml
pnpm-workspace.yaml		pnpm-workspace.yaml
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

OpenBrowser

What is OpenBrowser?

FAQ

Can ChatGPT be used for coding?

How to use ChatGPT for free coding?

How to use Claude for free coding?

Do I need an API key or paid AI subscription?

Is OpenBrowser a free alternative to API-based coding agents?

What is an AI harness?

Features

Requirements

Quick Start

Install pnpm

Installation

1. Install dependencies

2. Configure environment

3. Build

CLI Usage

Make openbrowser available in your terminal

Option A — Global command (recommended)

Option B — Run without global install

Option C — Interactive dev mode

Commands

Ask mode

Agent mode

Long prompts (file attachment)

Bridge Server

Start the server

Default URL

Endpoints

Verify the server is running

Browser Extension (Chrome)

Supported sites

Load the extension in Chrome

How the extension works

AI Response Format

Development

Project layout

Troubleshooting

openbrowser is not recognized (PowerShell)

Timed out waiting for browser AI response

Extension shows "Bridge server is not reachable"

Long prompt not attaching

Agent mode rejects the AI response

Changes not applied

Architecture

Contributing

License

About

Topics

Resources

License

Code of conduct

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Contributors

Uh oh!

Languages

Make `openbrowser` available in your terminal

`openbrowser` is not recognized (PowerShell)