zhikunqingtao/zhikuncode

What it solves

ZhikunCode is an open-source AI programming assistant designed to be self-hosted via Docker, allowing users to control the entire development workflow directly from a web browser on any device. It addresses the complexity of setting up AI coding environments by providing a unified interface for agent delegation, background task management, and direct integration with various LLMs, particularly those from Chinese providers.

How it works

The system utilizes a multi-layered architecture consisting of a Java Agent Runtime, a React-based browser interface, and a Python Capability Service for code analysis. It employs a "ContextCascade" engine with six levels of compression to handle extremely long conversations and a "SelfCorrectionLoop" to automatically diagnose and fix compilation or test failures. The assistant operates through a set of specialized agents (five roles) that can execute tasks in parallel, with a unified security gateway (Tool Gateway) that analyzes risks and manages permissions for all tool calls.

Who it’s for

It is aimed at developers who want a self-hosted, secure AI coding assistant that supports a wide range of LLMs (including Qwen, DeepSeek, Moonshot, and GLM) and provides a comprehensive workspace with integrated terminal, Git, and browser capabilities.

Highlights

• Browser-Based Control: Full workflow management, including permission approvals and task control, accessible via any browser without a client installation.
• Advanced Context Management: Six-layer compression and incremental folding to maintain performance in ultra-long dialogues.
• Coded-in-Java Core: High-performance core engine implemented in pure Java with zero external dependencies for REST API and WebSocket responses.
• Runtime Verification: A framework that uses JSONPath assertions and evidence chains (screenshots, logs, videos) to verify the output of AI-generated code.
• Multi-Modal Support: Intelligent routing for image analysis, automatically switching to vision-capable models when needed.
• Docker Deployment: Simple one-command setup via docker compose up -d for rapid deployment.