Developer Tools

Jcode.sh

Open-source AI coding agent that runs in your terminal — write code, run benchmarks, and coordinate agent swarms

jcode.sh

Verified Free tier

Quick facts
What is it	Open-source AI coding agent that runs in your terminal — write code, run benchmarks, and coordinate agent swarms
Pricing	Freemium — from $10/mo
Free tier	Yes
Platform	Desktop Application
Best for	automating code optimization tasks, running AI-powered code generation in terminal
Domain registered	2026

Data updated July 27, 2026

What does Jcode.sh do?

jcode is an open-source terminal coding agent written in Rust. It's a harness that sits on top of large language models and tries to squeeze as much useful work out of them as possible. You give it a coding task in your terminal, and it autonomously writes, tests, and optimizes code. It's not just a chat interface — it's an agent that can run commands, edit files, and iterate on its own output. The project is fully open source and publishes all its research publicly.

What makes jcode stand out is how it structures an agent's work. It uses a todo system where the agent rates its confidence in each task item — if confidence jumps too fast, the harness forces the agent to double-check its work. Every goal gets a "hill-climbability" score, and if that score is low, the harness pushes the agent to reframe the task into something measurable. jcode also supports swarms: multiple agent instances working in parallel on the same repository, coordinated by the harness. And it has an "auto-poke" feature that prevents the agent from declaring victory too early — if todos are still incomplete, the harness keeps poking. The page includes detailed benchmarks comparing jcode against other agents like Claude Code, showing significant speedups on tasks like float-print (398x vs 144x).

jcode is built for developers and AI researchers who want more control and transparency over how AI coding agents work. If you're tired of black-box assistants that stop after one attempt, jcode gives you a harness that pushes for persistence, measurability, and continuous improvement. It's especially useful for code optimization tasks, running custom benchmarks, and experimenting with agent swarms. Since it's open source, you can inspect every prompt, every tool call, and even contribute to the harness itself.

Developer Tools Development — Other LLM Testing

Key features

What makes it stand out

Confidence stepping — forces the agent to verify its work when confidence spikes too quickly, improving pass rates

Hill-climbable goals — every task gets a quantifiable score so the agent can iterate and improve over time

Swarm coordination — runs multiple agent instances in parallel on the same repository for faster exploration

Auto-poke — prevents early exits by checking the todo list and poking the agent when tasks remain incomplete

Uncontaminatable benchmarks — custom benchmark design that measures optimization depth instead of pass/fail

Who is Jcode.sh for?

Who benefits most from this tool

automating code optimization tasks

running AI-powered code generation in terminal

benchmarking AI coding agent performance

Pricing

Free tier available — start without a credit card

Free

Full jcode, no feature gates
Bring your own API keys
OAuth with your existing Claude / OpenAI subscriptions
Local models via Ollama / LM Studio
Open source, MIT licensed

Plus

$10.0/month

$18 of hosted model usage every month
Access to every Plus-tier model in the live catalog
Cheaper than ChatGPT Plus, built for agentic coding
Automatic failover across providers
Zero setup, no API keys to manage
Cancel anytime

Pro

$20.0/month

$40 of hosted model usage every month
Access to every Plus-tier model in the live catalog
For everyday agentic coding across active projects
Automatic failover across providers
Zero setup, no API keys to manage
Cancel anytime

Max

$100.0/month

$225 of hosted model usage every month
Access to every Plus-tier model in the live catalog
For heavy daily use and longer agent runs
Higher limits for sustained workloads
Automatic failover across providers
Zero setup, no API keys to manage

Ultra

$200.0/month

$500 of hosted model usage every month
Access to every Ultra-tier model, including Fable 5
For high-volume, multi-session workflows
Very high limits for swarms and parallel agents
Automatic failover across providers
Zero setup, no API keys to manage

Solo

$1000.0/month

$3,000 of hosted model usage every month
Access to every model in the live catalog
For power users and teams burning serious tokens
Priority routing and the highest limits
Swarm and multi-session heavy workloads

Trust & presence

Domain Domain registered 2026

Gallery

Click any image to enlarge

Alternatives in Developer Tools

opencode Developer Tools

Open source AI coding agent that works in your terminal, IDE, or desktop app.

n8n

Blackbox AI Code Developer Tools

AI coding assistant with autonomous agents — get real-time code suggestions, debugging help, and task automation.

Nous Research Developer Tools

An open-source, autonomous AI agent that runs on your server, learns from its tasks, and works across messaging apps.

Cycode Developer Tools

AI-powered application security platform that protects code from development to runtime with automated risk detection and remediation

PenguinHarness Developer Tools

Open-source framework for building and recursively self-improving AI agents

Agent AFK Developer Tools

Run coding agents autonomously, with remote supervision and verification, so you don't have to babysit the terminal.

Zcode Developer Tools

A Mac app for building native iOS and macOS apps, integrating AI agents and development tools into a unified workspace.

AutoCode Developer Tools

AI-powered platform that creates and manages autonomous coding agents to execute software development tasks.