BharatCode

The open source AI coding agent for VS Code, powered by bharatcode.ai.

BharatCode is an AI coding agent that meets you where you work — in VS Code. It's open source and MIT licensed.

Fork notice: BharatCode is a fork of Kilo Code (itself a fork of OpenCode), rebranded and reconfigured for the bharatcode.ai provider. The upstream projects and their contributors did the heavy lifting — this fork exists to bring their agent to the BharatCode platform. All credit to them; MIT licensed, attribution preserved. You pick from two models — qwen-3.8-27b (262k context) and deepseek-v4.1-flash (1M context) — through a single API key from bharatcode.ai, and every request goes through the BharatCode API at https://bharatcode.ai/api/model/v1.

Why BharatCode

BharatCode exists to make the bharatcode.ai setup effortless in VS Code. There is nothing to configure before you start:

1. Install the .vsix from the Releases page.
2. Paste your API key when prompted on first launch.
3. Chat.

That is the whole setup. What you get behind those three steps:

• Two models, one endpoint. qwen-3.8-27b (262k context) and deepseek-v4.1-flash (1M context) both come through one OpenAI-compatible endpoint at https://bharatcode.ai/api/model/v1 — switch mid-task from the model picker, no other provider accounts needed.
• No account juggling. One API key from bharatcode.ai; no separate gateway sign-in, no Kilo account or login flow — the cloud coupling from the upstream fork is stripped out of this build.
• No system-wide CLI to install. The extension carries its own backend binary (see How it works); nothing to put on your PATH.
• Your key stays local. It lives in the extension backend's auth store, never in settings.json, and is only sent to the provider (see PRIVACY.md).

How it works

The extension is not a thin API client. It runs a complete agent stack locally inside VS Code:

• Bundled backend binary. The extension ships its own bharatcode binary (extension/bin/bharatcode, .exe on Windows) inside the VSIX. It never calls a system-installed CLI.
• Lazy local server. On first use — your first chat submit, opening the Agent Manager, or an autocomplete prewarm — the extension's ServerManager spawns bharatcode serve --port 0. The OS picks a free port and the extension parses it from the child's stdout. That one child process is reused until it exits; if it crashes, the status chip in the chat header offers Restart server.
• Loopback-only auth. The server listens on 127.0.0.1 only and is protected by HTTP Basic auth with a random per-session password (32-byte hex) injected as KILO_SERVER_PASSWORD; the username is kilo. Nothing on your network can reach it without that password.
• Request path. The chat UI is a SolidJS webview that talks to the extension host over postMessage. The extension host calls the local backend over HTTP with server-sent events for streaming responses, using @kilocode/sdk. The backend forwards to the BharatCode API.

VS Code webview (SolidJS)
        │  postMessage
        ▼
Extension host
        │  HTTP + SSE (Basic auth, 127.0.0.1)
        ▼
Local backend  —  `bharatcode serve --port 0`  (bundled binary)
        │  HTTPS
        ▼
https://bharatcode.ai/api/model/v1
        ├── qwen-3.8-27b      (262k context)
        └── deepseek-v4.1-flash  (1M context)

Configuration is split in two places. VS Code settings (kilo-code.*) control UI and editor behaviour — themes, keybindings, display options. Agent runtime behaviour — model, permissions, instructions — lives in project and global agent config (kilo.json, ~/.config/kilo/). Both are covered in one guide: docs/CONFIGURATION.md.

Installation

The extension ships as a platform VSIX from the Releases page (a VS Code Marketplace listing is planned):

code --install-extension bharatcode-vscode-win32-x64.vsix   # Windows
code --install-extension bharatcode-vscode-linux-x64.vsix   # Linux

Or drop the .vsix file into VS Code's Extensions: Install from VSIX... command. Releases cover Windows, Linux, macOS, and Alpine on x64 and arm64.

On first launch, BharatCode prompts you for an API key — create one at bharatcode.ai. That's the whole setup.

Configuration

Everything user-configurable lives in one guide: docs/CONFIGURATION.md — API key, models, default model selection, config files, and how to get inline autocomplete working with a bring-your-own-key provider.

Quick answers:

• Change your API key: Command Palette → BharatCode: Set API Key
• Switch models mid-task from the model picker in the chat header
• The key is stored by the extension backend, never in your settings.json

Agents

BharatCode ships with specialized agents you switch between depending on the task. You can also build your own custom agents.

• Code — the default. Implements and edits code from natural language.
• Plan — designs architecture and writes implementation plans before any code gets written.
• Ask — answers questions about your codebase without touching any files.
• Debug — troubleshoots and traces issues.

What it does

• Code generation from natural language, across multiple files.
• Inline autocomplete with ghost-text suggestions and tab to accept — served by a bring-your-own-key provider (BharatCode's two models are chat-only; see configuration).
• Self-checking so the agent reviews and corrects its own work.
• Terminal and browser control to run commands and automate the web.
• Skills, modes, and MCP servers defined in plain text to shape the agent per project.
• Two models with mid-task switching — qwen-3.8-27b (262k context) and deepseek-v4.1-flash (1M context) — so you can match latency and reasoning to the job.
• Agent Manager — multi-agent sessions in an editor tab (Command Palette → Agent Manager), with each session isolated in its own git worktree.

Contributing

Contributions are welcome from developers, writers, and everyone in between. Start with the Contributing Guide for environment setup, coding standards, and how to open a pull request. See RELEASING.md for the VS Code extension release process.

Please review our Code of Conduct before getting involved.

License

MIT. You're free to use, modify, and distribute this code, including commercially, as long as you keep the attribution and license notices. See LICENSE.

Repository documentation

Where to go next, depending on what you need:

Document	Covers
README.md	What the extension is, how it works, how to install it (this file).
AGENTS.md	Repo-wide rules for agents and humans working on this codebase: build, dev, and test commands.
CONTRIBUTING.md	Prerequisites, environment setup, coding standards, and how to open a pull request.
TESTING.md	Spinning up a local backend and exercising it with curl/fetch without rebuilding the extension.
REVIEW.md	What reviewers look for on pull requests — human or agent.
RELEASING.md	The release process: version bump, per-platform VSIX builds, checksums, tagging.
PRIVACY.md	What data the extension collects and sends, including how the API key is stored.
SECURITY.md	How to report vulnerabilities in this repository.
docs/CONFIGURATION.md	The full user configuration guide: key, models, config files, autocomplete.

FAQ

Where does BharatCode come from?

BharatCode is a fork of Kilo Code (v7.8.8), which is itself a fork of OpenCode — see the fork notice at the top. Upstream history, commit links, and the kilocode_change markers in the source preserve full attribution.

---

Repository GitHub · bharatcode.ai