1jehuang/jcode

Project overview

jcode is a performant coding agent harness designed for multi-session workflows with optimized RAM usage and fast boot times. It provides semantic agent memory, multi-agent swarm collaboration, and cross-harness session resume capabilities for developers working in terminal environments.

Project type

AI Agent · AI Coding

Use cases

Coding & Development

Deployment

Refer to project documentation

License

MIT

Best for

• Developers who need to scale multi-session workflows while optimizing RAM usage and boot times.
• Users who require semantic agent memory to recall relevant information without actively calling memory tools.
• Developers using local runtimes or subscription-backed OAuth provider flows for AI models.

Key capabilities

• Built for scaling multi-session workflows with optimized RAM usage and fast boot up times.
• Embeds each turn and response as a semantic vector to automatically recall relevant information without actively calling memory tools.
• Spawn two or more agents in the same repository to manage native collaboration and conflict resolution automatically.
• Allows the agent to enter a mode to modify its own source code, build, and test changes automatically.
• Provides a side panel for auxiliary information, mermaid diagrams, and info widgets that take up negative space on the screen.
• Works with subscription-backed OAuth flows and many provider integrations.
• Comes with instructions on how to set up Firefox Agent Bridge for browser automation.
• Adds file structure information to the grep return so the agent can infer more of the file without reading it.

Limitations and risks

• Agents do not like to commit in a dirty git state with active changes, and git was not built for multi-agent workflows.
• Weaker models can make subtle, breaking changes when using self dev mode.

Getting started

• Setup difficulty is documented as easy because the tool can be installed with a single curl command. The first success path is installing via curl -fsSL https://jcode.sh/install | bash, then launching by running jcode.

Evidence and sources

• GitHub project description: Coding Agent Harness
• README: [](https://github.com/1jehuang/jcode/releases)
• README: The next generation coding agent harness to raise the skill ceiling.
• README: jcode is built to be as performant and resource efficient as possible.
• README: Jcode embeds each turn/response as a semantic vector.

Reviewed at: 2026-07-18T05:00:02Z