Dirgha Code is a free AI coding agent and terminal coding assistant developed by Dirgha AI for the mcp platform. Built with TypeScript and running on Node.js, it integrates with a multitude of large language model providers, including local options. Distributed as an npm package, it offers a powerful set of built-in tools for file system operations, version control, and code analysis. This program is designed to assist developers and agentic builders by automating complex coding tasks and enhancing AI-driven workflows.

What does Dirgha Code connect to?

Dirgha Code connects to numerous large language model providers such as Anthropic, OpenAI, Gemini, and NVIDIA NIM. It also supports specialized providers like DeepSeek, Groq, Mistral, and Perplexity. For offline operation, it integrates with local models including Ollama and llama.cpp, requiring no API key. This broad connectivity allows users to choose from a variety of AI backends, adapting to specific needs or preferences. Its design as an AI coding agent within the mcp ecosystem enables interaction with compatible AI clients.

Core features and tools

Dirgha Code provides 18 built-in callable tools to assist with development tasks. These include file system operations like fs_read, fs_write, fs_edit, and fs_ls, alongside shell and git for command execution and version control. Code analysis tools like go_to_definition, find_references, hover_documentation, and document_symbols are available. The github tool extends capabilities by wrapping gh CLI for actions such as pr_create and issue_list. Other tools include browser, checkpoint, and cron.

Who is it for?

Dirgha Code targets developers and "agentic builders," a group that includes writers, researchers, creators, and hardware teams. It supports those who need to automate complex coding tasks or orchestrate AI agents for various projects. Its ability to connect to many LLM providers and local models makes it suitable for users with diverse infrastructure needs. Skip this if you require a coding assistant that does not rely on external LLM providers or local model runtimes.

How to get started

The server authenticates using an API key passed as an environment variable or via a setup wizard. For the broader Dirgha API, authentication involves generating API keys from console.dirgha.ai and including them in "Authorization: Bearer" headers.

Use cases

• Parallel task execution: "dirgha fleet launch "<goal>"" decomposes a goal into independent tasks, runs them in parallel in separate git worktrees, and merges the results.
• Comparative analysis: "dirgha fleet triple "<goal>"" runs three approaches to a problem and selects the best outcome.
• Sequential agent chaining: "dirgha fleet dag "step1..." "step2..." "step3..."" allows agents to build on cumulative context across multiple steps.

Dirgha Code offers a versatile AI coding agent experience, integrating with numerous LLM providers and providing a rich set of development tools. Its support for both cloud and local models, coupled with advanced agentic workflows, positions it as a valuable asset for automating and enhancing coding processes. The BYOK authentication model provides flexibility for users managing their API access.