AI Coding Agent
Your AI coding agent, in the terminal or on your desktop
CoDev Code reads your code, runs your tools, and works where you already do — a terminal UI, a desktop app, and sandboxes it can be left alone in. Same sessions either way, signed in to your organization's own AI gateway.
Or install the terminal app:
$ npm i -g codev-codeOne agent, wherever you work
The terminal app, the desktop app and a sandbox all run the same CoDev Code. They read the same config and the same history, so a session you start in one turns up in the others.
Terminal-native
A full TUI in the terminal you already have open. Type codev and you are working; codev run puts the same agent in CI.
Desktop app
macOS, Windows and Linux, with its own server built in. Projects, tabs and recent sessions down the side, and nothing to start alongside it.
Work in a sandbox
Hand a task to an isolated environment that reaches only the folders you mount into it, and walk away — there is nothing left for a permission prompt to guard.
CoDev Code — the terminal app
A terminal-first coding agent that reads your code, runs your tools, and asks before it changes anything.
- A TUI built for the keyboard — sessions, diffs, and the todo list above the prompt
- Multi-provider — pick the model that fits each task
- Real code awareness with LSP and tree-sitter
- Extensible with MCP servers, plugins, skills, and custom agents
- Scriptable —
codev runfor CI and automation - Shares its sessions, config, and credentials with the desktop app
CoDev Desktop — and Work
The same agent in a window, with a second side to it. Code is your machine. Work is a sandbox on it, where the agent can run unsupervised without reaching anything you did not hand it.
Code — your machine
Your projects on the local server, or on any WSL, remote or SSH server you add. Sessions run with the reach you have, and ask before anything destructive.
Work — your sandboxes
An isolated VM or container that reaches your disk only through the folders you mount, and the network only through the model gateway.
Keeps itself current
Auto-updating on all three platforms, and you decide how much of that happens without asking.
Your models, inside your network
Every session signs in with your corporate account and routes through the AI gateway you host, so prompts and code stay on your own infrastructure.
Gateway and SSO
Sign in once. The terminal app, the desktop app and every sandbox reach models through the gateway you run, and nowhere else.
Agents that know your code
codev codegraph enable indexes your repositories into a knowledge graph served over MCP — fewer tool calls, fewer tokens, better answers.
Get started in three steps
- 1
Install it
Download the desktop app, or install the terminal app from npm.
npm i -g codev-code - 2
Sign in
Authenticate with SSO and pick a default model. The desktop app asks the first time you open it.
codev configure - 3
Start working
Open the project you want worked on, and say what you want done.
cd /path/to/project codev