jcode - open-source AI coding agent for the terminal

jcode is an MIT-licensed terminal coding agent written in Rust, with durable memory, background tasks, parallel agent swarms, and broad model-provider support.

Visit Website
jcode - open-source AI coding agent for the terminal

Introduction

Overview

jcode is an open-source AI coding agent built in Rust for terminal-based development. It combines an interactive TUI with non-interactive runs, resumable sessions, persistent background operation, and the ability to coordinate multiple agents in parallel.

The project is MIT licensed, making the coding harness inspectable and customizable. Its design emphasizes parallel throughput, efficient resource use, and the tools, context, memory, and feedback loops surrounding the selected model.

Key Features

  • Runs many agent sessions in parallel through swarm workflows.
  • Keeps durable memory and supports named session resumption.
  • Handles background tasks through a persistent server with attachable clients.
  • Displays images, LaTeX, structured edits, status widgets, and file-conflict notifications in the terminal.
  • Loads repository instructions from AGENTS.md and machine-wide instructions from the user's home configuration.

Published measurements report about 10.4 MB of additional proportional memory per session and 48.7 ms to first input on the tested machine. These figures describe the project's own launch tests and may vary by environment.

Installation and Use

jcode provides installers for macOS, Linux, and Windows, a Homebrew option for macOS, and source builds through Rust tooling. Termux is also supported when its required compatibility packages are installed.

After installation, users launch jcode in a terminal and connect a provider during first use. Common modes include opening the TUI, submitting a single non-interactive prompt, resuming a previous session, or running a background server and connecting clients to it.

Models and Configuration

jcode supports subscription sign-in flows for Claude, OpenAI, Gemini, GitHub Copilot, Azure OpenAI, Ollama, and LM Studio. Named profiles cover several popular OpenAI-compatible services, while custom profiles can target other compatible endpoints, including local servers without authentication.

Configuration covers the default provider and model, reasoning and failover behavior, display preferences, feature toggles, keybindings, swarm routing, concurrency, and lifecycle hooks. Headless authentication options support SSH and other environments without a browser.