PRODUCT - CRUXY CODE

Cruxy Code. A coding agent for your terminal.

Early access, as a CLI. Cruxy Code reads your codebase, edits files, runs your tests and commands, and opens pull requests. By default it asks before every edit and command. IDE plugins and a standalone editor are planned.

Install Cruxy Code See pricing →

~/web

$▌

Interface preview. CLI output is real; model replies are illustrative.

Install

Install Cruxy Code.

Cruxy Code is in early access as a CLI. You need Node.js 20 or later, a Cruxy account, and an API key.

npm install -g @cruxy/cli
export CRUXY_API_KEY=cxy_live_...
cruxy run "explain this codebase"

Create an account at cruxy.ai, then create an API key in the app.
Set the key as CRUXY_API_KEY, or run cruxy login to save it. The first interactive run without a key walks you through setup.
To try it without installing, run npx @cruxy/cli run with your task.

Capabilities

What Cruxy Code does today, and what is planned.

In your terminal today

Cruxy Code is a CLI: cruxy opens an interactive session and cruxy run does one task. Planned: plugins for VS Code and JetBrains, and a standalone editor.

Reads your codebase first

It lists, reads, globs and greps your files, searches the codebase with an index built on your machine, and loads your project's CRUXY.md or AGENTS.md every session.

Four modes

Manual, the default, asks before every edit and command. Plan proposes a plan for your approval first. Auto-approve and full-auto let reversible actions run and still ask before irreversible ones. Switch with /mode or Shift+Tab. Planned: a pair mode with line-by-line review.

Real git workflow

cruxy pr writes a conventional commit on a new branch, pushes it and opens a GitHub pull request, all after one approval, and never commits to main or master. Planned: triggering your deploy pipelines.

Any language, any stack

It works through file tools and your shell, so it is not tied to a language or framework. Optional language servers cover TypeScript, JavaScript, Python, Go and Rust.

Runs your tests

It runs your test command, reads the failures and keeps fixing, stopping after four failing runs in a row. Failure output is parsed for Vitest and Jest. Planned: generating tests, and reviewing diffs before a pull request.

Repository memory

It remembers facts across sessions: project memory in .cruxy/memory, recalled once you trust it with cruxy memory trust, and personal memory in ~/.cruxy/memory.

Multi-file changes

One patch can create, update and delete files across the repository atomically, under a single approval. Planned: running your linter and formatter automatically after a change.

Diff-first review

Every edit shows its diff before you approve it. cruxy rollback restores the tree to before any turn, and /undo-last undoes the latest one. Planned: approving hunk by hunk.

Usage you can see

Every run ends with its token usage and cost. /usage, cruxy usage and cruxy limits show more.

How it works

Plan. Write. Ship.

1. Describe the work.

Run cruxy for a session, or cruxy run with a task: a feature, a refactor, a bug fix. It reads the codebase and works out the change.

2. Approve as it goes.

By default it shows each edit as a diff and each command before it runs, and waits for you. In plan mode you approve the plan first; the auto modes let reversible actions run.

3. Open the pull request.

cruxy pr commits on a new branch, pushes and opens a GitHub pull request after one approval. Planned: deploying once you have authorized that step.

Models

Runs on Kavi, Vaani and Mira, routed per request.

By default Cruxy Code sends auto, and the Cruxy API picks Mira, Vaani or Kavi for each request by its size and complexity. Pin a model with /model for the session, in .cruxy/config.json for a project, or with the CRUXY_MODEL environment variable.

Kavi

The deepest reasoning tier, for complex refactors and long-context work.

Learn about Kavi →

Vaani

The everyday tier: fast and capable, at a low cost per token.

Learn about Vaani →

Mira

The fastest tier, for short requests. Planned: inline autocomplete in your editor.

Learn about Mira →

Compare all three →

Who uses Cruxy Code

Solo or team. Side project or production.

Solo developers shipping side projects, weekend builds, and personal tools, with an agent that, by default, asks before it changes anything.

Professional engineers at companies who want to move faster on the parts of the job that aren't where the value is - boilerplate, refactors, test fixes, migrations.

Engineering teams sharing conventions through a committed CRUXY.md, so every session starts from the same rules.

Common questions

Quick answers about Cruxy Code.

How is Cruxy Code different from Cursor or Copilot?: Cursor is an editor; Copilot is an autocomplete plugin. Cruxy Code is an agent in your terminal: it plans multi-step changes, edits files, runs your tests and commands, and opens pull requests, and by default asks before each edit or command. Working inside your existing editor is planned.
Does Cruxy Code work with my private codebase?: Yes. It runs on your machine and sends the conversation, including the output of the tools it runs, such as files it reads, to the Cruxy API. There is no repository upload. Your prompts and outputs are not used to train Cruxy's models without your consent. As in Cruxy chat, the API can save durable facts from a session to your account's memory.
Which languages and frameworks does Cruxy Code support?: It is not tied to a language: it reads and writes any text file and runs your toolchain through your shell. There is no official language list. Test failure parsing covers Vitest and Jest, and optional language servers cover TypeScript, JavaScript, Python, Go and Rust.
Can Cruxy Code commit to my repo without my approval?: No. In the default mode it asks before every edit and command. cruxy pr asks once for its whole plan (branch, commit, push, pull request) and never commits to main or master. Force-pushes and other irreversible actions always ask, even in the auto modes. Planned: standing approval per repository or per branch.
Which plans include Cruxy Code?: Every plan, from Free to Enterprise: Cruxy Code is in early access and is not gated by plan. With an API key, usage is metered per token; signed in with cruxy login, it counts toward your plan's usage limits. Team features for Code, such as shared repository context, are planned.
Where does Cruxy Code run?: In your terminal, on any machine with Node.js 20 or later. Install it with npm install -g @cruxy/cli. Plugins for VS Code and JetBrains, and a standalone editor, are planned.

Pricing

Cruxy Code is in early access on every plan.

Every plan - Free, Core, Advanced, Pro and Enterprise. Cruxy Code is not gated by plan.
API key - A key from cruxy.ai is metered per token.
cruxy login - Signing in from the CLI counts toward your plan's usage limits.

See all plans →

What Cruxy Code does

Plan it, approve it, undo it.

Choose how much it asks: /mode plan proposes a plan before anything runs.

~/web

Interface preview. CLI output is real; model replies are illustrative.

You stay in control: Cruxy Code proposes, you approve.

~/web

Interface preview. CLI output is real; model replies are illustrative.

Changed your mind? /undo-last restores the files from before the last run.

~/web

Interface preview. CLI output is real; model replies are illustrative.

Features

Built for the terminal, and for trust.

CLI first

Lives in your shell: run cruxy for a session, or cruxy run for one task. No app to switch to.

Git-aware

Knows your current branch and whether the tree is dirty, and lists changed files when it needs them.

Approval by default

Asks before every edit and every command, with a diff for each edit. Irreversible actions always ask.

Any language

Works through file tools and your shell, so it is not tied to a language or framework.

Runs your tests

Runs your test command and works through the failures. Planned: generating tests.

Routed per request

By default the Cruxy API picks Mira, Vaani or Kavi for each request. Pin one with /model.

Who it's for

For engineers who ship.

Solo founders and indie hackers

Hand Cruxy Code the boilerplate and the boring refactors, and approve each change as it lands.

Backend engineers shipping fast

Multi-file refactors as one patch, with the diff in front of you before any file changes.

Teams onboarding to new repos

Ask Cruxy Code to walk you through an unfamiliar codebase: it reads and searches the files itself. Planned: generating tests for legacy modules.

FAQ

Common questions.

Does Cruxy Code see my entire codebase?

It reads the files it needs with read-only tools (list, read, glob, grep and a search index built on your machine), and those reads do not ask for approval. Edits and commands do. Nothing is uploaded wholesale: what it reads goes to the Cruxy API as part of the conversation.

Can I use Cruxy Code without internet?

No. Every model turn goes to the Cruxy API. Only the codebase search index runs on your machine. A local-only mode for sensitive enterprise repos is planned.

Does Cruxy Code work with my IDE?

Not yet. Cruxy Code is a CLI today, so it runs in any terminal, including the one in your editor. Plugins for VS Code and JetBrains, and a standalone editor, are planned.

How is it different from Claude Code or Cursor?

Same category as Claude Code: an agent in your terminal. Cruxy Code runs on Cruxy's own models, Kavi, Vaani and Mira, with plans billed in rupees.

What's the pricing?

Cruxy Code is in early access on every plan, from Free to Enterprise, and is not gated by plan. With an API key from cruxy.ai, usage is metered per token. Signed in with cruxy login, it counts toward your plan's usage limits.

Other Cruxy products

Cruxy

General-purpose AI assistant

Cruxy Cowork

Autonomous desktop coworker

Cruxy Bench

Web chat with Cruxy Code

Cruxy Guard

Automatic pull request review

Build faster. Ship cleaner.

Cruxy Code, in your terminal today. In early access on every plan.

Install Cruxy Code See pricing →