Section 1
What jcode actually is
jcode is a coding agent that lives in your terminal, like Claude Code or Codex. You talk to it, it reads your files, writes code, runs commands, and fixes its own mistakes.
The difference is what it's built out of. The big tools wrap the AI model in a layer of JavaScript that handles your prompts, tools, and memory. That layer is heavy and slow to start. jcode rebuilt that whole layer in Rust, the language people reach for when something has to be fast. So it barely touches your computer's memory and starts almost instantly.
It's free and open source (MIT license), so anyone can read it, run it, and change it. You don't pay jcode anything. You bring your own AI model, usually the Claude or ChatGPT subscription you already have.
One honest note: it's a real terminal tool. You'll paste a command or two. You do not need to know how to code to use it, and the next section hands you a prompt that does the install for you.
Section 2
Why it beats the big tools
These are the maker's own benchmarks from the jcode README. The point isn't the exact decimal, it's the size of the gap. Built in Rust instead of JavaScript, jcode runs light enough that the numbers stop being close.
RAM per added session
jcode ~9.9 MB · Claude Code ~212.7 MB
Time to first frame
jcode 14 ms · Codex CLI 882.8 ms
Time to first frame
jcode 14 ms · Claude Code 3,436 ms
RAM, 10 sessions open
jcode 117 MB · Claude Code 2,300 MB
Here's the real unlock hiding in those numbers. Because each agent is so light, you can run a whole swarm of them in the same project at once, all working together and catching each other's mistakes. Try that with the heavy tools and your machine chokes. More on the swarm below.
Section 3
Install it
Two ways to do this. The easy way is to let an AI agent install it for you. The manual way is one line.
The easy way: let an agent do it
Paste this into any coding agent you already have open (Claude Code, Cowork, Codex, or jcode itself). It detects your system, installs jcode, reuses a login you already have, and runs a test. No guessing which command is right for your machine.
Copy-paste setup prompt
Set up jcode on this machine for me.
1. Detect the operating system, available package managers, and shell environment, then install jcode using the best matching command below instead of pointing me somewhere else:
- macOS with Homebrew:
brew tap 1jehuang/jcode
brew install jcode
- macOS or Linux via install script:
curl -fsSL https://raw.githubusercontent.com/1jehuang/jcode/master/scripts/install.sh | bash
- Windows PowerShell:
irm https://raw.githubusercontent.com/1jehuang/jcode/master/scripts/install.ps1 | iex
2. Verify that `jcode` is on my PATH.
3. Launch `jcode` once to confirm it starts.
4. Before any interactive login, check what I already have set up and reuse it. Look for existing credentials and config such as:
- Claude: ~/.claude/.credentials.json, ANTHROPIC_API_KEY
- OpenAI/ChatGPT: ~/.codex/auth.json, OPENAI_API_KEY
- Gemini: ~/.gemini/oauth_creds.json
- GitHub Copilot: ~/.config/github-copilot/
5. Prefer whichever provider is already configured and confirm it works with `jcode auth-test --all-configured`.
6. Only if nothing is configured, walk me through the single login I need (e.g. `jcode login --provider claude`).
7. Run a smoke test with `jcode run "say hello"` and confirm it works.
8. Explain any step that still needs me (a browser login, an API key, a device code). The manual way: one line
Open your terminal and paste the line for your system.
macOS & Linux
macOS with Homebrew
brew install jcode
Windows (PowerShell)
Section 4
Use the subscription you already pay for
This is the part that saves you money. jcode doesn't charge you and it doesn't force you onto a pay-per-use API. It logs into the same Claude Pro or ChatGPT Plus subscription you already have, so you get a fast new agent on top of a plan you're already paying for.
Pick the model you pay for and run its login line. It opens a browser, you click approve, done.
Claude
jcode login --provider claude ChatGPT / Codex
jcode login --provider openai Gemini
jcode login --provider gemini GitHub Copilot
jcode login --provider copilot Out of tokens on one account? Type /account to switch to a second one mid-session. jcode also supports Gemini, OpenRouter, DeepSeek, local models through Ollama and LM Studio, and a long list of others, but Claude or ChatGPT is all most people need.
Section 5
It remembers, with zero setup
Most agents forget everything the second you close them. jcode has memory built in, and you don't configure a thing.
As you work, it quietly turns each exchange into a searchable note and files it in a memory graph. Next time something related comes up, it pulls the relevant memory back in on its own, the way a person remembers a detail mid-conversation. No memory commands to run, no tokens wasted constantly re-reading old notes.
Every so often it cleans the whole thing up in the background, merging duplicates and clearing out stale or conflicting notes so the memory stays sharp instead of bloating over time. If you ever want to drive it yourself, there are tools to search or save a memory by hand, plus a search across every past session.
Section 6
The swarm: many agents, one project
This is the feature the heavy tools can't really pull off, and it's why the low memory matters. Open two or more jcode agents in the same project and they automatically start working as a team.
When agent A edits a file that agent B is in the middle of, the server taps agent B on the shoulder so the code never shifts under its feet without it knowing. The agents can message each other directly, DM one teammate, or broadcast to everyone in the repo, and conflicts get resolved on their own.
An agent can even spin up its own swarm. Hand it a big job and it becomes the coordinator, spawns a few workers to split the work in parallel, tracks who's done what, and reports back. That's a small team of agents on one codebase, running on your laptop, for free.
Section 7
Your first commands
Once it's installed and logged in, these are the four worth knowing. Run the first one and you're in.
jcode Launch the full terminal app (TUI).
jcode run "say hello" Run one command without opening the app.
jcode --resume fox Jump back into a past session by its memorable name.
jcode serve Run it as a background server, then attach more agents to it.
That's the whole thing. Free agent, your own subscription, memory and a swarm out of the box. The repo lives at github.com/1jehuang/jcode if you want to read the source or watch the demos.
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.