jcode: AI coding agent running on 27.8MB RAM

jcode is an AI coding agent optimized to the extreme for memory usage and boot speed. It uses 27.8MB of RAM per single session, which is up to 13 times less than existing tools such as Claude Code or OpenCode. The time to first frame rendering is 14ms, providing instant response compared to other tools that consume hundreds to thousands of milliseconds.

It features a built-in memory system that embeds conversation content into semantic vectors and stores them in a graph structure. The agent automatically recalls relevant information without token consumption and verifies information retrieval via side agents when necessary. Memory is automatically extracted and consolidated at session end or when semantic drift occurs, maintaining long-term memory.

It provides Swarm functionality where the server automatically manages conflicts when running multiple agents in the same repository. When one agent modifies a file, other agents receive immediate notifications to review or ignore the differences. Agents can autonomously create sub-agents for parallel task processing when needed, with the main agent acting as a coordinator.

It supports subscription-based OAuth flows for Claude, OpenAI, Gemini, GitHub Copilot, and others, allowing users to continue using their existing AI models. The terminal UI includes a dedicated library that renders mermaid diagrams 1800 times faster, and a side panel allows real-time viewing of file contents or diffs. It is suitable for developers who want a resource-light environment when scaling multi-session workflows.