Back to stories
Tools

Moonshot's Kimi Code Brings Open-Source Agentic Coding to the Terminal

Michael Ouroumis2 min read
Moonshot's Kimi Code Brings Open-Source Agentic Coding to the Terminal

Moonshot AI, the Beijing-based startup behind the Kimi assistant, has open-sourced Kimi Code — a terminal-native coding agent that can autonomously navigate codebases, write and debug code, and execute multi-step engineering tasks directly from the command line.

What Kimi Code Does

Unlike IDE-integrated copilots, Kimi Code is designed for developers who live in the terminal. It operates as a CLI tool that accepts natural-language instructions and translates them into sequences of file edits, shell commands, and git operations. The agent can explore unfamiliar repositories, understand project structure, run tests, and iterate on failures without human intervention.

Moonshot claims that Kimi Code achieves a 62 percent pass rate on the SWE-bench Verified benchmark, putting it within striking distance of proprietary leaders like Claude Code and Codex while being fully open-source under an Apache 2.0 license.

Architecture

The system is built on Moonshot's Kimi-Dev model, a 72-billion-parameter code-specialized model fine-tuned for agentic workflows. The model supports 128,000-token context windows, enabling it to ingest large files and maintain coherent plans across multi-step tasks.

Kimi Code's agent loop follows a plan-execute-verify pattern: it first analyzes the task and relevant code, generates an execution plan, carries out the steps, and then verifies the result by running tests or linting. If verification fails, it automatically attempts to diagnose and fix the issue.

Why It Matters

The release is significant for several reasons. First, it demonstrates that open-source coding agents are rapidly closing the gap with proprietary alternatives. Our comparison of Claude Code, Copilot, and Cursor showed that proprietary tools still lead on complex tasks, but Kimi Code's SWE-bench score suggests that gap is narrowing fast. Second, it provides a foundation that other developers and companies can build on, customize, and deploy without API dependencies or usage fees.

"Terminal-based coding is where the most productive developers work," said Moonshot CTO Yang Zhilin. "We wanted to meet them where they are, not force them into a new editor."

Getting Started

Kimi Code is available now on GitHub with installation via pip. It supports Python, JavaScript, TypeScript, Go, Rust, and Java out of the box, with community-contributed language support expected to expand quickly.

Early community feedback has been positive, with developers praising the tool's ability to handle real-world repository structures rather than just isolated coding puzzles. GitHub's Agent HQ has already added Kimi Code as a supported agent option. For a comprehensive comparison of terminal-based AI tools, see this Claude Code vs Copilot CLI vs Gemini CLI guide. Several contributors have already submitted pull requests adding support for additional languages and custom tool integrations.

Learn AI for Free — FreeAcademy.ai

Take "Prompt Engineering Practice" — a free course with certificate to master the skills behind this story.

More in Tools

Google Turns Chrome Into an AI Coworker With Auto Browse, Powered by Gemini 3
Tools

Google Turns Chrome Into an AI Coworker With Auto Browse, Powered by Gemini 3

At Cloud Next 2026, Google unveiled Auto Browse, a Gemini 3-powered agent inside Chrome that handles multi-step web tasks for consumers and enterprise Workspace users.

5 days ago3 min read
OpenAI Launches Workspace Agents, Retires Custom GPTs for Teams
Tools

OpenAI Launches Workspace Agents, Retires Custom GPTs for Teams

OpenAI today unveiled workspace agents in ChatGPT as a research preview, positioning them as a direct replacement for custom GPTs and pitching Codex-powered shared agents at Business, Enterprise, Edu, and Teachers customers.

6 days ago2 min read
Cloudflare Launches Agent Memory Private Beta to Give AI Agents Persistent Recall
Tools

Cloudflare Launches Agent Memory Private Beta to Give AI Agents Persistent Recall

Cloudflare's new Agent Memory service extracts and stores information from AI agent conversations so models can recall context across sessions without bloating the token window, addressing one of agentic AI's biggest bottlenecks.

1 week ago2 min read