cleankoda logo

Agentic coding with true human-in-the-loop control

From issue tracker to pull request. cleankoda captures full project context to deliver clean code directly in your terminal.

$ Installation bash / zsh
$ curl -fsSL https://cleankoda.com/install | bash
Prerequisites: Git, Docker & Python 3.10+
// ABOUT THE AGENT

What is cleankoda?

Cleankoda is a collaborative, TUI-based coding agent that integrates seamlessly into your entire development cycle—from ticket to pull request.

Agentic coding only delivers sustainable productivity gains when AI and humans operate as a well-coordinated team. Instead of writing code in isolation, cleankoda understands the holistic context of your work: the agent pulls user stories and acceptance criteria directly from issues (e.g., in Jira), plans the implementation with you in the terminal, and executes the coding in controllable, incremental steps.

Based on the HULA principle (see illustration), cleankoda combines an understanding of your requirements with maximum human decision-making power—resulting in better code, fewer iteration loops, and full control within your familiar terminal environment.

cleankoda Agent Workflow Diagram
HULA Workflow Loop (Task Setup ➔ Planning ➔ Coding ➔ PR) from Takerngsaksiri et al. "Human-In-the-Loop Software Development Agents"
// CORE ARCHITECTURE

Designed for greater productivity, control, and data sovereignty

[01 / TRACKER]

Integration into the software development workflow

Connect directly to Jira Cloud/Server, Trello boards, and GitHub Issues. Fetch tickets, update status, and attach patch context automatically.

Jira Trello GitHub Issues
[02 / PERFORMANCE]

Terminal-Native TUI Performance

Instant startup, sub-millisecond keyboard responsiveness, and low memory footprint. Built for developers who live in zed, vs code, and neovim.

Keyboard-Driven zed vs code
[03 / SAFETY]

Safe Execution & Unified Diffs

No black-box code overwrites. Preview interactive side-by-side or unified diffs, run local test suites in sandbox, and approve changes granularly.

Interactive Hunks Local Sandbox Line-by-Line
[04 / AGNOSTIC]

Model-Agnostic Engine

Use local LLMs via Ollama & vLLM for 100% offline privacy, or connect to OpenAI, Anthropic, and Gemini models with custom API keys.

LiteLLM Ollama vLLM OpenAI / Anthropic
// PROJECT STATUS AND ROADMAP

Dreaming in Code

cleankoda is currently an experimental concept study and not yet intended for production use. The ecosystem surrounding autonomous coding agents (such as OpenCode, Antigravity, or Claude Code) is evolving at a breathtaking pace. As a solo developer with limited time, it is hardly possible to keep up with every daily trend—yet the core vision remains alive: a collaborative coding agent based on genuine clean-code principles.

The project is continuously being developed. If you’d like to help shape the future of cleankoda, contribute your own ideas, or submit code, feel free to check out the GitHub repository, open an issue or pull request, or reach out to me directly. Any and all support is very welcome!