Command Code

Command Code is a terminal-based AI coding agent that learns your coding taste from accept/reject/edit signals. It supports individual, team, and API-oriented plans, with model choice and local storage of taste and memory.

AI Code Assistant AI Agent Development AI Developer Tools Programming Agent

Visit Website

Overview

Command Code is an AI coding agent for developers who work in the terminal and want the assistant to adapt to their coding style over time. The site positions it as a frontier coding agent that ships, fixes, tests, and refactors while continuously learning from the user’s accept, reject, and edit decisions.

Its core workflow centers on taste-1, a meta neuro-symbolic model that distills those signals into project-level skills and personal memory. The product also supports interactive and headless modes, shared taste across teammates, and a CLI-first installation flow, making it a fit for individuals, teams, and API builders who want a coding agent that follows existing conventions instead of imposing new ones.

Core features

Continuous taste learning

Command Code learns from every accept, reject, and edit, then turns those signals into project-level skills and personal memory so later sessions follow the patterns you prefer.

Multiple agent modes

The product supports multiple working modes, including interactive CLI use, headless automation, plan-only analysis, and a background sandbox for heavier tasks.

Built-in developer tools

The agent includes shell access, file operations, grep, extended thinking, and related tooling so it can work across code changes without leaving the terminal.

Project memory and extensibility

It supports custom `/agents`, persistent `/memory`, reusable `/skills`, `/commands`, MCP servers, plugins, and session sharing through `npx taste push/pull`.

Multi-directory workspace support

The features page describes multi-directory workspaces, where sibling repos, shared packages, or infra folders can be added as first-class workspace roots.

Model flexibility with local privacy

The pricing and features pages describe support for open-source and premium models, including taste-1 and other listed providers, with no training on your code.

Common use cases

Coding inside an established project
Use Command Code when you want an AI pair programmer that learns the naming, structure, and conventions of a specific codebase instead of starting from generic defaults.
Automated or unattended coding tasks
Use the headless and background modes for tasks like refactors, fixes, or automation runs that do not require constant interaction.
Team standards and shared conventions
Use the sharing workflow to push project taste to teammates or pull shared taste into another repo so a team can converge on the same conventions.
Cross-repo and monorepo work
Use the multi-directory workspace support when work spans a monorepo, sibling repo, or shared infra folder and the agent needs context from more than one root.
API-based coding workflows
Use the API-oriented plan if you are building against OpenAI- or Anthropic-compatible clients and want usage-based billing with provider docs.

Pros and Cons

Pros

Learns from real editing actions instead of relying only on prompts.
Works from the terminal and includes both interactive and headless workflows.
Supports project memory, reusable skills, and team sharing through `npx taste push/pull`.
Offers multiple model choices, including open-source and premium options.
States that code is not used for training and taste storage stays local.

Cons

The site does not provide a full public integration catalog, so supported tools and workflows are only described at a high level.
Some practical details, such as precise plan limits and usage boundaries, are presented on pricing pages rather than summarized in a simple product matrix.

FAQ

What is Command Code?

Command Code runs as a terminal-based coding agent and continuously learns from accept, reject, and edit actions. The site describes it as a frontier coding agent that ships, fixes, tests, and refactors from the terminal while adapting to your coding taste.

Which models can I use with it?

The site says it supports the models listed in its docs, including Anthropic, OpenAI, Google, xAI, DeepSeek, Qwen, Kimi, GLM, MiniMax, and more. The pricing page also says the API plan is compatible with OpenAI and Anthropic clients.

Can I bring my own API keys or use it with an API client?

Yes. The pricing page offers a Provider plan for API builders that uses OpenAI or Anthropic-compatible endpoints and supports bring-your-own-key behavior. The site also says new vendors are added regularly.

Does Command Code train on my code?

No. The site states that your code, skills, and memory stay on your machine and that Command Code never trains on your work.

How do teams share coding taste?

Teams can share taste with `npx taste push` and `npx taste pull`, and the pricing page says team plans use pooled usage, shared billing, and central admin.