GitHub

Runs on your own computer · Free to use

A team of AI helpers that builds, checks and asks first.

Tell OpenCodingPartner what you want in plain words. A planner, a coder and a reviewer work together on it. The reviewer catches mistakes, the coder fixes them, and nothing changes until you say yes.

  • Ada Planner
  • Linus Coder
  • Grace Reviewer
  • Sentinel Safety check
  • Atlas Virtual computer

The demo is a read-only tour. To use OpenCodingPartner for real you need Node.js 22.15 or newer and either a free model running on your computer (through an app like Ollama) or your own AI key.

Illustration: you ask for a bakery website. Ada the planner makes a plan. Linus the coder builds it in a separate copy of your project. Grace the reviewer finds two problems: prices in dollars instead of rupees, and a phone number left as YOUR_NUMBER. Linus fixes both, Grace checks again, and then Studio asks you to approve saving the changes.

Illustration of how a request moves through the team, not a live session.
Works with apps that run free models on your computer, and with AI providers you have a key for
  • Ollama
  • LM Studio
  • vLLM
  • OpenRouter
  • Gemini
  • Anthropic
  • OpenAI
  • Sarvam

One real request, start to finish

Recorded in Studio on a local install, using a hosted model through OpenRouter. Clicks play at real speed; time spent waiting for the AI is sped up. Click a clip to watch it.

Like a small software team that checks its work and asks before it changes anything.

Everything below works today when you install OpenCodingPartner on your own computer.

Teamwork

Bots check each other’s work

A planner splits up the job, a coder builds it, and a reviewer checks it. If the reviewer finds a problem, the coder fixes it and they go round again before anything reaches you.

Self-check

Checks its own files before saying “done”

Before it reports back, the bot looks over what it made again.

  • !A page it promised is missing
  • !Leftover placeholders like YOUR_NUMBER
  • !$ where you asked for ₹
You’re in charge

Approvals in plain words

When a bot wants to do something that matters, it asks you in everyday language.

Virtual computers

A computer of its own for each bot

A bot can get its own virtual screen with a real web browser, to open your app and click around like a person would. You can watch it work, or press Take over and use it yourself.

For now this needs Linux or Windows with WSL2, plus TigerVNC, xdotool, ffmpeg, XFCE and Chromium (one install command).
Safe to try

Changes happen in a copy

In a Git project, bots edit a separate copy (a worktree). You see exactly what changed, then keep it or undo it. Without Git, each edit is asked about first.

Your choice of AI

Free local model, or your own key

Run a free model such as Gemma 4 on your computer, using an app like Ollama, LM Studio or vLLM. Or connect OpenRouter, Gemini, Anthropic, OpenAI or Sarvam with your own key.

Privacy

Passwords kept locked

Website passwords go into an encrypted vault on your computer; AI keys stay in a private settings file only your account can read. Bots can also run routines and remind you later.

Four steps, and you stay in charge.

  1. Install it

    Run one command on your computer. It opens a page called Studio in your web browser.

  2. Pick an AI model

    Choose a free model running on your computer, or paste an API key from a provider you already use.

  3. Ask in plain words

    Describe what you want built or fixed. The bots make a plan, build it, and check it.

  4. Review and decide

    Look at the changes. Approve them, or undo them with one click.

100 small web apps, built on one computer

Games, to-do lists, shop catalogues, calculators and more. 93 of the 100 were written by OpenCodingPartner’s coder bot with the free Gemma 4 model, running offline in Ollama; the other 7 come from its built-in templates. Open any of them in your browser.

Free on your own computer

With a model running on your own computer, it costs nothing. If you use a paid AI provider with your own key, that provider bills you directly.

Starter: we host the AI

Coming soon. Starter is not on sale yet, and this site has no checkout. When it opens, you buy it inside the app. Details may change before then.

₹999 / 30 days, GST included
  • For people who don’t want to run a model or manage API keys. No setup.
  • Two models: Starter · Fast and Starter · Smart
  • Includes about $7 of hosted AI usage for 30 days
  • Unused credit ends after 30 days. No automatic renewal: top up when you need more.
  • Local models and your own keys stay free

The online demo is a read-only tour of a sample Studio. It is free and does not take payment.

Try the read-only demo

The technical details

What is in the box today, and how it fits with the tools you already use.

Bots, worktrees and review

  • A lead bot breaks a goal into tasks and hands them to coder, reviewer and security bots; the reviewer can send work back to the coder until it passes.
  • Before reporting done, the bot re-checks its own output for missing files and leftover placeholders.
  • Each task runs in its own Git worktree under .ocp/worktrees/<task-id>; your working tree is not touched until you merge.
  • Review diffs hunk by hunk, then merge or roll back.
  • Studio binds to loopback only: 127.0.0.1:4317.

Models

  • Local: any model you run in Ollama, LM Studio or vLLM (for example Gemma 4).
  • Hosted, with your own key: OpenRouter, Gemini, Anthropic, OpenAI, Sarvam, or any OpenAI-compatible endpoint.
  • No default model — you choose one on first run.

Code graph

  • Indexes symbols, imports and call edges for TypeScript, JavaScript, Python, Go and Rust into a local SQLite graph (.ocp/graph.sqlite).
  • Tools such as code_get_callers and code_blast_radius show what a change could affect before you merge.

Virtual computers, vault and sign-in

  • Per-bot virtual desktops run locally on Linux or WSL with TigerVNC, xdotool, ffmpeg, XFCE and Chromium installed, with mouse, keyboard and browser control and a Take Control button.
  • Secrets live in an encrypted local vault.
  • Google and Microsoft OAuth using your own OAuth app credentials.
  • Ready-made connectors (Slack, Telegram, GitHub, Railway and more), plus routines and reminders.
Available

ocp app

Studio, the web interface on 127.0.0.1:4317.

Available

ocp mcp

An MCP server exposing 130+ tools to MCP-capable editors and agents.

Available

ocp tui

A keyboard-driven terminal interface.

Coming soon

Editor, desktop & mobile

A VS Code extension, a desktop app, and Android / iOS apps are in development and not in any store yet.

Common questions

What do I need to use it?
A Mac, Linux or Windows (WSL2) computer with Node.js 22.15 or newer and Git. The bots’ virtual computer screens need Linux or WSL2 for now. You also need an AI model: either a free model such as Gemma 4, running on your computer in an app like Ollama or LM Studio, or an API key from a provider like OpenRouter, Google Gemini or Anthropic.
Does it cost anything?
Running it on your own computer is free. If you use a paid AI provider, that provider bills you directly for what you use. Starter, a paid option where we host the AI for you (₹999 for 30 days), is coming soon and not on sale yet.
Does my code leave my computer?
Studio runs on your computer. If you use a model running on your computer, your code stays there. If you use an online AI provider, the parts of your code the bots work on are sent to that provider.
Can a bot break my project?
When your project is a Git repository with at least one commit, bots make their changes in a separate copy (a Git worktree). Nothing reaches your main copy until you approve it, and you can roll changes back.
Is there a phone app or a VS Code extension?
Not yet. They are coming soon, but are not in any app store today. Today you use Studio in your web browser, the terminal interface, or the MCP server.
What is the license?
OpenCodingPartner is source-available, not open source. The Studio, CLI and wrappers use the Elastic License 2.0: free for personal use and inside your business, but not to resell as a hosted service. The core engine is proprietary and free to run on your own computer. The 100 showcase app templates are MIT. The LICENSE file has the full terms.

Install on your computer

First install Node.js 22.15 or newer and Git. OpenCodingPartner is not a standalone program; it runs on Node.js. Then pick one of these:

Terminal
npx github:bgagandeep/opencodingpartner-community app --port 4317

When it is running, open http://127.0.0.1:4317 in your browser. Studio will ask you to choose an AI model.

Open your local Studio

Studio runs on your own computer once OpenCodingPartner is installed and started. No account is needed.

Not installed yet?