Leading Chinese AI company Moonshot AI has officially launched its dedicated programming tool: Kimi Code. The release marks Kimi's transition from a "general assistant" to a "professional coding tool" in the field of developer productivity.

image.png

Full Scenario Coverage: From Terminal to Mainstream Editors

Kimi Code demonstrates strong compatibility and flexibility, aiming to cover the entire developer workflow:

Independent Operation: Supports direct invocation in terminal environments, meeting the needs of quick debugging and script processing.

Seamless Plugin Integration: Achieves comprehensive integration with mainstream IDEs and editors such as VSCode, Cursor, JetBrains, and Zed, allowing developers to enjoy AI empowerment without changing their existing habits.

image.png

Technical Core: Driven by the K2.5 Multimodal Model

Differing from traditional text-assisted programming, Kimi Code fully leverages the multimodal advantages of the K2.5 model:

Visual Interaction: Supports programming assistance by directly inputting images or videos, greatly simplifying visual-driven development tasks such as UI replication and chart-to-code conversion.

Ability Migration: The system can automatically identify the user's existing skill stack and efficiently migrate it into new programming paradigms or workflows.

Performance Upgrade: In internal benchmark tests called Kimi Code Bench, which cover the full range of tasks including building, refactoring, and testing, the new model with K2.5 achieved a qualitative breakthrough compared to the previous version.

Embracing Open Source: Launching Agent SDK at the Same Time

To build an AI programming ecosystem together, Moonshot AI also open-sourced the Agent SDK behind Kimi Code.

Highly Customizable: Developers can create personalized intelligent agents (Agents) based on this SDK.

Case References: The official has already uploaded multiple example demos on GitHub, allowing developers to quickly get started and explore.

With the addition of Kimi Code, domestic AI programming tools are accelerating into a new phase of deep integration with large models, offering global developers more competitive intelligent options.