Have you been fixing AI code that never looks like yours? Command Code learns your taste, so the fixes stop
Command Code is a terminal AI agent whose taste-1 layer records every accept, reject, and edit you make, then distills the signal into project-level skills. 29,000 developers, 320 releases shipped, $1/mo Go plan with 4× credits on DeepSeek V4 Pro. When the fixes stop, you know it worked.
Context from: Commandcode | Commandcode | Commandcode
The decision it puts on your desk
Install Command Code Go ($1/mo) by end of day. Pick one workflow you already know the failure modes of. Run the agent on it for 5 days. Count the corrections you still make to its output. If the count drops below your current IDE agent on the same task, extend taste-1 to a second workflow by end of month. If it does not, you have spent $1 to find out the harness is not for you. The decision is whether taste learning fits your workflow, and the only way to know is to run the loop.
Command Code is a terminal AI coding agent with 29,000 developers and 320 releases shipped. The thing most people miss: it watches your accept and reject decisions, then stops making the mistakes you keep fixing.
What it does
A coding agent in your terminal. You give it a task. It runs. It ships features, fixes bugs, runs tests, refactors code, audits design, recolors interfaces. Seventeen modes ship out of the box, including /design for interface audits and /review for code review. The default loop is interactive. Run it headless with command-code -p for scripts and CI. Run it in the background with command-code --yolo when you trust the harness. Run it in a sandbox when the work touches production.
The full developer toolchain is in: file ops, shell, grep, extended thinking, agent orchestration, persistent /memory, custom /agents, reusable /skills, /commands, MCP servers, and a plugin system. The framework is built to be hacked.
The taste loop
The differentiator is taste-1. Every time you accept, reject, or edit a diff, the agent records the signal. The signal gets distilled into a project-level skill, a markdown file that captures a convention. "pnpm over npm." "Tabs over spaces." "Prefer vitest." The next session opens with the convention already in place. No rules to write. No prompts to maintain.
The team version shares the skills. npx taste push publishes a project skill. npx taste pull installs one. Skills are open files in the repo, reviewed in PRs like any other code. The first team to standardize on a shared taste library moves faster than the team that does not.
Models and pricing
Every model in the docs ships out of the box: Anthropic, OpenAI, Google, xAI, DeepSeek, Qwen, Kimi, GLM, MiniMax, and more. The $1/mo Go plan comes with $10-40 in free credits that stretch 4× on DeepSeek V4 Pro, 2.7× on MiniMax M3, 99% off on MiMo V2.5 Pro. Pro and Team plans add seats, more compute, and shared taste registries. Your code never trains their model. Privacy policy says it. The architecture confirms it.
Install
npm i -g command-code. Sign in. Start coding.
← Previous
OpenCode ships 6 tab primitives in 9 days, then rebuilds the desktop around them
Next →
Command Code raised $5M and shipped taste-1, the first coding agent that learns from every accept, reject, and edit
The Morning Recap
We sift the day. You make the calls.
The decision on every story, in your inbox before your first call.