Skip to content
Get packs

Guide · Open source

How to use jcode: install and run parallel coding agents

Updated · written with AI from the project's own documentation and checked against our catalogue data

jcode is an open-source coding agent you run from the terminal. It uses AI models to read, edit and work on code inside a project, and it is designed to run several agents at once with low memory use. This guide shows you how to install it, start it, and use its swarm features, based on the project README.

jcode1jehuang/jcode · ★ 21.4k · MIT

Quick start

  1. 1Run the install command. Paste the macOS and Linux install command into your terminal and let it finish.
  2. 2Open a terminal in your project. Change into the repository you want the agent to work on before starting jcode.
  3. 3Start jcode. Open jcode from the terminal in a new session and confirm it responds to typed input.
  4. 4Connect a provider. Sign in with a supported provider, such as Claude or OpenAI Codex via OAuth, using the steps in the provider docs at jcode.sh/docs.
  5. 5Spawn a second agent. Start another agent in the same repo so the server can manage file-change notices between them.
  6. 6Review diffs when notified. When the server says a file you read has changed, decide whether it matters, and check the diff if it might conflict.

Install jcode on macOS or Linux

The README gives one install command for macOS and Linux. Run it in your terminal, then check that the install worked by opening a new terminal session and starting jcode.

The README lists no prerequisites and does not describe a Windows install method, so this section only covers macOS and Linux.

curl -fsSL https://jcode.sh/install | bash

Understand what jcode is best suited for

jcode is a terminal (TUI) tool, not a graphical editor. It is built for developers who are comfortable with command-line tools and want several agents working on one repository at the same time.

It is a poor fit for non-programmers, because it edits source code, and for teams that need a verified security or sandboxing model, because the README does not describe one.

  • Good fit: developers running several agents on one repo
  • Good fit: people concerned about RAM and startup speed
  • Good fit: users who want to connect existing Claude or OpenAI Codex accounts via OAuth
  • Poor fit: non-programmers
  • Poor fit: anyone who needs a graphical interface

Run several agents in the same repo (swarm)

The README says that when you spawn two or more agents in the same repo, the server manages them so they can collaborate. If agent A edits a file that agent B has read, the server notifies agent B. Agent B can ignore the notice or check the diff for conflicts.

Agents can message one agent directly, broadcast to all agents on the server, or message only the agents working in that repo. The README also says agents can spawn their own helper agents. In that case the main agent acts as coordinator and the spawned agents act as workers. This can run headless or with the interface open.

  • Spawn two or more agents in the same repo to get automatic file-change notices
  • Use direct messages or broadcasts to coordinate agents
  • Let an agent spawn workers for parallel tasks

Connect your AI provider accounts

The README says jcode supports a wide range of providers, including Claude OAuth and OpenAI Codex OAuth. If you already have a Claude or OpenAI Codex account, you can connect it this way instead of managing separate keys.

The README links to the provider list at jcode.sh/docs. It does not give the exact sign-in steps, so check that page for the current setup process.

Check the performance claims yourself

The README publishes benchmarks comparing jcode with OpenCode, pi, Claude Code and Codex CLI. It reports RAM per additional headless session and time to first input across 10 interactive launches. The benchmark script is linked in the README, so you can run it on your own machine.

These numbers come from the project's own README. Test them on your hardware and workload before relying on them.

  • RAM test: scripts/bench_headless_memory.py
  • Startup test: time until typed text appears, 10 launches

Explore the self-development mode

The README describes a self-dev mode. You can tell a jcode agent to enter self dev mode, and it will modify its own source code. The README says it can edit, build and test its own code, then reload its binary and keep working. Because this lets an agent change the tool itself, read the README and docs before using it on important work.

What people use it for

  • Running two or more agents on one repository so they can see each other's file edits
  • Splitting a large task into parallel jobs, with one agent coordinating workers
  • Running many headless sessions on a machine where RAM is limited
  • Working on a project in the terminal with an agent that edits source files

Frequently asked questions

How do I install jcode?

The README gives one install command for macOS and Linux: curl -fsSL https://jcode.sh/install | bash. Run it in your terminal and then start jcode from a new session. The README does not list a Windows install method.

Is jcode free?

jcode is released under the MIT licence, according to the README. The MIT licence allows free use, modification and distribution, but check the repository's licence file for the full terms. Any model provider you connect may charge its own fees.

What are the requirements for jcode?

The README lists no prerequisites. It does say jcode runs in the terminal and that you need an account with a supported provider to use AI models. Check the project docs for any system requirements.

What is jcode's alternative?

Projects the README compares it with are OpenCode, Claude Code, Codex CLI and pi. Each is a coding agent, but they differ in memory use, startup time and features. Try the one that fits your workflow and compare the results on your own project.

Does jcode support Claude and OpenAI Codex accounts?

Yes, the README says jcode supports Claude OAuth and OpenAI Codex OAuth, along with a wide range of providers. You can use existing accounts instead of creating new API setups.

Is jcode safe to run on my code?

The README does not describe a verified security or sandboxing model. Because jcode edits source code, review its changes before committing them, and avoid using it where you need a verified security boundary.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

jcode suits developers who are comfortable in the terminal and want several AI coding agents working on one repo, especially if RAM and startup speed matter to them. Non-programmers, people who want a graphical editor, and teams that need a documented security or sandboxing model should choose a different tool.

Install options & downloads