In the AI coding tool space dominated by Claude Code and OpenAI Codex, an open-source Rust project is rapidly catching up.
jcode (github.com/1jehuang/jcode) is a Coding Agent Harness—it’s not another wrapper, but an orchestration layer between the user and the underlying AI model, managing prompts, tool calls, memory, and parallel sessions. The project has 1.3K stars, 2,811 commits, with the latest commit just 53 minutes ago, showing extremely high development activity.
Why We Need Agent Harness
Agent Harness is the core architectural layer of AI coding tools. Claude Code and Codex CLI each implement their own harness, but they don’t interoperate. jcode’s approach: make this layer open-source, replaceable, and extensible.
This means you can:
- Switch between different LLMs at the bottom layer (Claude, GPT, open-source models)
- Extend capabilities through skill plugins
- Manage multiple parallel Agent sessions
- Unified memory and context management
Core Features
The repository structure reveals jcode’s design ambition:
- crates/ - Rust multi-crate architecture with high modularity, including complete code quality fix swarms
- .jcode/skills/optimization/ - Built-in optimization skills supporting Agent self-improvement
- ios/ and figma/ - Mobile support in development, including wgpu scene preview
- mockups/jcode-mobile/ - Mobile UI design
The project particularly focuses on self-development scenarios—recent commits include fixes for “avoiding self-dev rustc triggering earlyoom kills,” showing the project is building itself.
Quick Start
# Clone the project
git clone https://github.com/1jehuang/jcode.git
cd jcode
# Build (requires Rust toolchain)
cargo build --releasePrimary Sources: