jcode: a high-performance coding agent harness with semantic memory and collaborative multi-agent swarms

What it solves

jcode is a high-performance coding agent harness designed to increase the "skill ceiling" for AI-assisted development. It addresses the resource inefficiency and slow boot times common in many CLI-based AI agents, while providing advanced memory systems and collaborative multi-agent workflows (swarms) for complex software projects.

How it works

• Semantic Memory: The system embeds every turn and response as a semantic vector. It uses a graph of memories and cosine similarity checks to automatically recall relevant information without needing explicit tool calls or consuming excessive tokens.
• Memory Management: A background "memory sideagent" extracts and stores memories based on semantic drift or turn count, while an "ambient mode" periodically consolidates and cleanses the memory graph.
• Swarm Intelligence: Users can spawn multiple agents in a single repository. The server manages these agents, notifying them of file changes made by others and providing messaging channels (DMs or broadcasts) to resolve conflicts and coordinate tasks.
• High-Performance UI: It features a custom rendering engine capable of over 1,000 FPS, a dedicated side panel for real-time file updates and diffs, and a custom terminal (Handterm) to enable smooth, partial line scrolling.
• Provider Integration: It supports a wide array of providers via OAuth and OpenAI-compatible endpoints, including Claude, OpenAI, Gemini, GitHub Copilot, and local runtimes like Ollama and LM Studio.

Who it’s for

Developers who need a resource-efficient, highly customizable coding agent that can handle multi-session workflows and complex, collaborative AI-driven development across large codebases.

Highlights

• Extreme Efficiency: Significantly lower RAM usage and faster "time to first frame" compared to other major AI coding CLI tools.
• Autonomous Swarms: Agents can autonomously spawn their own teammates to parallelize tasks, turning the main agent into a coordinator.
• Integrated Visuals: Native rendering of Mermaid diagrams and specialized "info widgets" that utilize negative screen space.
• Flexible Provider Support: Extensive built-in support for both subscription-backed OAuth flows and self-hosted/local LLM endpoints.