Getting Started with CodeMantis

Prerequisites

Before installing CodeMantis, make sure you have:

• macOS 12 (Monterey) or later, or Windows 10/11 (64-bit) — CodeMantis is a native desktop application for macOS and Windows.
• At least one coding agent CLI installed — Claude Code (see Installing Claude Code), OpenAI Codex (see Installing Codex), or OpenCode (see Installing OpenCode). You can install several — having Claude Code and Codex together unlocks per-task routing — but any one alone is enough to use CodeMantis. You do not need a terminal or npm to install them; see In-app install & sign-in below.
• An active subscription for the agent(s) you install — Claude Pro ($20/mo) or Max ($100-200/mo) for Claude Code, or ChatGPT Plus/Pro/Business for OpenAI Codex.

Not sure which agent to pick? See Choosing Your Coding Agent.

Download and Install

Downloads come straight from CodeMantis and always point to the latest published version. You can also grab either build from the download page.

macOS

1. Download the latest CodeMantis for macOS (.dmg) — a universal build for Apple Silicon and Intel.
2. Open the .dmg and drag CodeMantis into your Applications folder.
3. Launch CodeMantis from Applications. On first launch, macOS may ask you to confirm you want to open the app — click Open.

Windows

1. Download the latest CodeMantis for Windows (.exe installer). For managed or silent deployments, use the .msi package instead.
2. Run the installer and follow the prompts.
3. If Windows SmartScreen appears, click More info then Run anyway. Launch CodeMantis from the Start menu.

In-app install & sign-in

You do not need to open a terminal or install Node/npm before using CodeMantis. On first launch, the Welcome Screen checks your environment and can set up its coding-agent CLIs for you directly from the Requirements card — CodeMantis needs at least one of Claude Code, OpenAI Codex, or OpenCode installed and ready.

• Install or update a CLI — When an agent is missing or outdated, an Install / Update button appears (in a yellow callout box for Claude Code, inline on the row for Codex and OpenCode). Click it and CodeMantis runs the agent's official native installer in the background, streaming the live install output beneath the button. When it finishes, the Requirements card re-checks automatically. Expand Show details / advanced to see the exact command it runs — or the legacy npm install -g … command if you would rather use npm.
• Sign in to a CLI — Click the Sign in button on an authentication row. A sign-in overlay opens with a live terminal and starts the agent's browser OAuth flow (Claude Code runs /login; Codex runs codex login). Complete the login in the browser window that opens, then close the overlay — the row flips to Logged in on its own. OpenCode has no interactive sign-in; instead you add your own model-provider API keys under Settings → Agents → OpenCode.

CodeMantis is ready as soon as at least one agent is installed and signed in, and you can add the others later. For per-agent details, see Installing Claude Code, Installing Codex, and Installing OpenCode — or Choosing Your Coding Agent if you are not sure which to pick.

Opening a Project

When CodeMantis launches, you can start working in two ways:

• Open an existing folder — Click Open Project and select any folder on your Mac. CodeMantis will detect its structure and initialize a new session. When both Claude Code and Codex are installed, an Agent Picker appears so you can choose which agent powers the session.
• Create from a template — Choose from built-in project templates (Next.js, Tauri, Python, and more) to scaffold a new project with sensible defaults.

Sending Your First Message

Once a project is open, the chat panel is front and center. Type a message describing what you want to build or change — for example:

Add a login page with email and password fields

Your active agent will analyze your project, propose changes, and ask for approval before modifying any files.

Session Modes at a Glance

Claude Code sessions use the Mode Selector:

• Normal — You approve each action before it runs. Best for learning and sensitive work.
• Auto-Accept — The agent works without pausing for approval. Best for fast iteration.
• Plan — Reasoning only, no code changes. Best for architecture discussions.

You can switch modes at any time with Command + . or by using the mode selector in the chat panel.

Codex sessions use the Codex Policy Pill instead — two independent dials for sandbox (read-only / workspace-write / danger) and approval policy (untrusted / on-request / on-failure / never). Same goal, different model.

Learn more in Session Modes Explained.

Next Steps

• Understanding the Three-Panel Layout — Learn how the workspace is organized.
• Choosing Your Coding Agent — Claude Code vs. Codex, plus per-task routing.
• Session Modes Explained — Decide how much control to give your agent.
• Installing Claude Code / Installing Codex — Troubleshoot CLI setup issues.
• Per-Task Agent Routing — Route different task categories to different agents (v1.5.0).
• Setting Up AI API Keys — Configure API keys for changelogs, assistants, and SpecWriter.