Matcha CLI · Available now

MatchaCode

From an idea to code that runs.

A calmer way to ship code.

Read the repo, plan the change, write the feature, and run the tests from your terminal. Bring the models you want.

curl -fsSL https://cli.matchacode.net/install.sh | bash

macOS and Linux. Runs locally. Bring your own models.

Explore example projects
AI-generated architecture website mockup with ivory typography, a courtyard house photograph, and a project index.
Projects & possibilities

Projects & possibilities

Explore what you could build

A website with character. A clearer view of your data. A small tool that feels just right. Start with what you want to make.

AI-generated architecture website mockup with ivory typography, a courtyard house photograph, and a project index.
01Brand website

Give good work room to breathe

Natural light, considered materials, and editorial typography put the work first.

Example taskBuild a responsive portfolio for an architecture studio.

AI-generated analytics dashboard mockup with visitor metrics, a slate-blue trend line, and traffic charts.
02Business application

Find clarity in your data

Bring trends, traffic sources, and key metrics into one considered view.

Example taskAdd trend analysis and date filters to a dashboard.

AI-generated mobile booking website mockup with a studio photo, dates, available times, and a confirmation summary.
03Mobile website

Make every step feel simple

From choosing a day to confirming a time, a calm flow for a smaller screen.

Example taskRefine a booking flow and its submission states for mobile.

AI-generated JSON utility mockup showing input, formatted output, and validation status in a dark interface.
04Developer tool

A little tool. A better workflow.

Input, format, and check. A focused interface for doing one thing well.

Example taskBuild a JSON formatter and validator.

The workspace

One request. A complete workflow.

Read, plan, edit, and test. Follow a task from the first request to the final check, all in your terminal.

Illustrative workflow
Read the demonstration

Add rate limiting to the API and cover it with tests.

Read

Reading src/api/route.ts, src/api/route.test.ts, and AGENTS.md

Plan

  1. Wrap the handler in a rate-limit layer
  2. Add tests. Keep the existing assertions.

Change

- export const POST = handler+ export const POST = withRateLimit(handler)

Test

npm test

Rate limiting added. 12 tests passed.

Features

What it can do.

Capability illustrations

  1. Fanned paper planes connected by a green path, illustrating the structure of a repository.
    src/
    index.tsformat.ts
    01

    Read the repo

    Read files and search the codebase to see how the project works.

  2. Two offset paper sheets with a green insertion, illustrating a precise code edit.
    format.ts
    return name;return name.trim();
    02

    Write features and edit code

    Change files directly. Edits appear as inline diffs.

  3. Three paper steps leading to a glass gate, illustrating a plan awaiting approval.
    1. Read
    2. Plan
    3. Approval
    Awaiting approval
    03

    Plan before editing

    When the path is unclear, read first and propose a plan. Code waits for your say-so.

  4. An array of paper tiles with one displaced tile, illustrating a failing test to repair.
    format.test.ts
    Failed Passed
    04

    Fix tests, inspect failures

    Run commands and tests, inspect the result, then keep working from the failure.

  5. Two interchangeable architectural modules on a continuous track, illustrating model choice within a session.
    Model AModel B
    Same session
    05

    Switch models

    Switch by model name during a session, using the models your configured Provider offers.

  6. Paper guides and tool modules connected to a graphite work surface, illustrating project rules and tools.
    Project
    AGENTS.mdMCPSkills
    06

    Bring rules and tools

    Read project rules and connect the tools you use through MCP and skills.

Models

Use the models you want.

Bring an API key for xAI, OpenAI, or Anthropic. Or connect Ollama and other compatible endpoints.

MatchaCode is not a model provider. These names are examples of closed, open, local, and aggregator endpoints. They are not a native Provider catalog, and they are not endorsement.

Promise

What stays yours.

Sessions and rules stay on your machine. Inference goes to the Provider you choose.

Local storage & model requests · Illustrated

Your machine

Matcha CLI

~/.matcha
  • Sessions
  • Rules
Model requests
Model responses

Your models

Your chosen Provider

Cloud API or local Ollama

Your permissions

Edits and commands follow your permission settings.

Bring yours

Your rules can come with you.

Matcha CLI reads the conventions other assistants already use. You do not start over.

  1. 01

    Reads the files as they are

    AGENTS.mdCLAUDE.mdCursor rules
  2. 02

    Preview Claude config, then import

    /import-claude

    PreviewImport

  3. 03

    Continue a Cursor, Claude, or Codex session

    /resume-cursor/resume-claude/resume-codex

CLI

Work in the terminal.

  1. 01

    Work in the repo

    Attach files with @. Diffs, commands, and task lists share one conversation.

    matcha "fix the failing auth test and run it"
  2. 02

    Pick up where you left off

    /resume opens a previous session. matcha -c continues the latest one. You can also rewind a turn.

    matcha -c
  3. 03

    Plan, then edit

    When the approach is ambiguous, read and propose. Code waits for your say-so.

    /plan
  4. 04

    Switch models in-session

    Select a model offered by your configured Provider without leaving the workflow.

    /model grok-4.5

Surfaces

The CLI is available now.

  • CLI

    Available now

    Read the repo, edit files, and run commands in the terminal.

    Download
  • IDE Plugin

    Coming soon

    The same agent, in your editor.

  • Desktop App

    Coming soon

    A desktop client, later.

  • Mobile App

    Coming soon

    Continue the same workflow on your phone, later.

Start

Get started.

Download the CLI, pick a model, and start. macOS and Linux are the primary platforms.

curl -fsSL https://cli.matchacode.net/install.sh | bash

MatchaCode itself is free for now. There is no account and no product cloud. Model usage is billed by your Provider.