CodeMantis — The desktop app Claude Code, Codex & OpenCode deserve
CodeMantis is a free, fair-source native desktop application for macOS and Windows that wraps the Claude Code, OpenAI Codex, and OpenCode CLIs in a three-panel graphical interface — pick which agent each session uses, or route different tasks to different agents. Built with Tauri v2 and Rust on a pluggable agent architecture, it launches in under 3 seconds, uses under 120MB of memory, and includes AI-powered spec writing, 10+ project templates, integrated terminals, and a live preview browser. Under 50MB download, licensed under the Fair Core License.
Native · macOS & WindowsBuilt with Tauri v2 + RustSee it in action
CodeMantis in under 3 minutes
Three Coding Agents. One Desktop App.
CodeMantis is the native desktop app that treats Claude Code, OpenAI Codex, and OpenCode as equals. Pick your agent when you open a project. Switch mid-flight. Or route different task categories to different agents — write specs with Claude, run hot main chat on Codex, keep experiments on an open or local model via OpenCode. The agent you bring is yours; the interface is the same.
- Agent Picker in the Project Picker when more than one CLI is installed
- Per-session agent selection — sessions carry an agent_id
- Per-task routing for main chat, SpecWriter, Help, and the Assistant panel (v1.5.0)
- 7-day usage-split card shows tokens and cost per agent
- Adaptive controls — Mode Selector for Claude, Codex Policy Pill for Codex, a native Plan pill for OpenCode
- OpenCode runs open & local models over ACP — OpenRouter, Ollama, LM Studio, or the free OpenCode Zen gateway
- Bring your own vendor subscription (Claude Pro/Max, ChatGPT Plus/Pro/Business) or your own model provider — no lock-in
- Pluggable AgentAdapter — additional CLIs ship as a config, not a rewrite
Write the Spec, Not the Code
Your coding agent implements what you describe — but the quality of the description determines the quality of the result. SpecWriter is an AI conversation partner that draws out the details you'd forget to mention: loading states, error handling, validation rules, edge cases. It produces professional specification documents, verification checklists, and multi-session implementation plans. Then Self-Drive implements everything automatically through Claude Code or Codex — from idea to finished feature.
- Interactive multi-turn conversation
- Attach mockups, screenshots, and PDFs
- Reads your existing codebase and references real file paths
- Confidence tagging — the AI tells you when it's guessing
- Generates companion verification checklists
- Implementation Guide breaks specs into multi-session plans
- Self-Drive automates the entire implementation — build, verify, fix, advance
- Two modes: spec a new app from scratch, or add a feature to an existing project
See Everything at Once
A coding agent's power comes from its ability to read, write, and execute across your entire project. CodeMantis makes that visible. Chat conversations flow on the left. Every file read, write, edit, and bash command appears in the Activity Feed on the right — color-coded, expandable, with one-click approval. For Claude Code sessions, pick Normal mode (approve each action), Auto-Accept (work freely), or Plan (reasoning only). For Codex sessions, the Codex Policy Pill lets you dial sandbox × approval-policy independently.
Start Projects in Seconds
Your agent is at its best when it understands your project from the first message. CodeMantis gives it that head start. Pick from 10+ curated templates — Next.js, Vite+React, FastAPI, Astro, Expo, and more. One click scaffolds the project, installs dependencies, and generates a CLAUDE.md (or AGENTS.md) file that teaches the agent the project's structure, conventions, and commands. From empty folder to productive conversation in under a minute.
Preview Your App Live
A native preview window shows your running application right alongside your coding conversation. Auto-detects your dev server, supports responsive viewport presets — phone, tablet, desktop — and captures console logs so you can spot errors without switching to Chrome DevTools. It's the workflow you always wanted: describe a change, watch your agent implement it, see it live.
More Than One Brain
Sometimes you want a second opinion before committing to an approach. Open parallel AI assistants — the latest models from OpenAI, Google, or Anthropic — in tabs alongside your main coding session. Brainstorm architecture. Review code. Draft documentation. Compare suggestions. Each assistant has per-session cost tracking. The Assistant panel is agent-agnostic: it works the same whether your main session runs Claude Code or Codex. Your coding agent stays in control of your files; the assistants are your advisory board.
Everything in One Window
CodeMantis is a native desktop app built with Tauri v2 and Rust — not an Electron wrapper, not a web page in a frame. It launches instantly, uses minimal memory, and respects platform conventions on macOS and Windows: native menus, keyboard shortcuts, proper window management. Inside that native shell: a Monaco code editor, xterm.js terminals, MCP server management for 15+ services, AI-powered changelogs, slash commands, and six carefully designed themes. Everything a developer expects, nothing a developer would compromise on.
What the Community Says
Early feedback from developers using CodeMantis with Claude Code.
I went from a terminal-only Claude Code workflow to CodeMantis and immediately felt the difference. Seeing the activity feed separate from the chat is a game-changer — I actually know what Claude is doing to my files now.
Early Beta Tester
Full-Stack Developer · Discord community
SpecWriter alone is worth the download. I used to spend 20 minutes writing prompts for Claude Code. Now I have a conversation with SpecWriter and get a proper spec in 5 minutes that Claude follows precisely.
Community Member
Product Engineer · Discord community
Finally a Claude Code interface that respects macOS conventions. It launches fast, uses minimal memory, and the three-panel layout makes so much more sense than a single scrolling terminal.
Community Member
macOS Developer · Discord community
Frequently Asked Questions
Common questions about CodeMantis and how it works with Claude Code and Codex.
CodeMantis is a free, fair-source native desktop application for macOS and Windows that wraps Anthropic's Claude Code CLI, OpenAI's Codex CLI, and OpenCode (open and local models via ACP) in a graphical interface. You pick which coding agent each session uses — or route different task categories (main chat, SpecWriter, Help, Assistant) to different agents from Settings > Agents. Three-panel layout with a file tree, chat, and activity feed, plus integrated terminals, a Monaco editor, AI-powered spec writing, and a live preview browser. Built with Tauri v2 and Rust, it launches in under 3 seconds and uses under 120MB of memory.
Yes. The CodeMantis core is free — and it will stay free. There are no API keys required for Claude Code or Codex, and no telemetry. CodeMantis is licensed under the Fair Core License (FCL-1.0-ALv2): you can read, use, modify, and redistribute it for anything except building a competing product, and every version becomes open source (Apache 2.0) two years after its release. You do need a way to reach at least one of the three supported agents — a Claude Pro ($20/mo) or Max ($100-200/mo) subscription for Claude Code, a ChatGPT Plus/Pro/Business subscription for OpenAI Codex, or a model provider for OpenCode (OpenRouter, a local Ollama/LM Studio endpoint, or the free OpenCode Zen gateway). A paid Pro tier with advanced automation features is planned for later this year; the core product stays free, with no token markup — you always bring your own agent subscription.
Both CLIs run in a single scrolling pane where chat messages and tool operations are interleaved. CodeMantis separates them into dedicated panels: chat on the left, activity feed on the right. You also get a built-in file editor (Monaco), integrated terminals, a live preview browser, project templates, AI-powered spec writing (SpecWriter), MCP server management, and per-task agent routing — all in one native desktop window. The same panel separation makes sense for Claude Code, Codex, or OpenCode; the agent you pick is just a setting.
Cursor and VS Code extensions add AI to an existing code editor. CodeMantis takes the opposite approach: it builds a purpose-designed interface around a coding agent as the primary workflow — and it isn't locked to one. You get a dedicated activity feed showing every file operation, SpecWriter for creating implementation-ready specifications, session modes (Normal, Auto-Accept, Plan) for Claude, the Codex Policy Pill (sandbox × approval-policy) for Codex, or a native Plan pill for OpenCode, and a native desktop app that uses 40-120MB of memory versus 300-500MB for Electron-based editors.
CodeMantis runs on macOS 12 (Monterey) or later (Apple Silicon or Intel) and Windows 10 or 11 (64-bit). You also need at least one of the supported agent CLIs installed: Claude Code (npm install -g @anthropic-ai/claude-code) with a Claude Pro/Max subscription,, OpenAI Codex (npm install -g @openai/codex, or brew install codex) with a ChatGPT Plus/Pro/Business subscription, or OpenCode (curl -fsSL https://opencode.ai/install | bash) pointed at a model provider. You can install several — CodeMantis will let you pick per session or route per task. The app itself is under 50MB to download.
SpecWriter is an AI-powered specification assistant built into CodeMantis. Instead of giving your coding agent a vague prompt, SpecWriter guides you through an interactive conversation to extract every detail — loading states, error handling, validation rules, edge cases. It produces a structured specification document with a verification checklist, then hands off to whichever agent you've chosen for spec_writer in Settings > Agents. Two modes: New Application (from scratch) and Feature Mode (for existing projects, with codebase context).
Yes. CodeMantis includes Multi-AI Assistant tabs where you can run OpenAI, Google Gemini, or Anthropic API models alongside your active session. These assistants are chat-only (they cannot edit files or run commands) — perfect for brainstorming, code review, or getting a second opinion. The Assistant panel is agent-agnostic: it works the same whether your main session is Claude Code, Codex, or OpenCode. Each assistant tracks per-session token usage and cost.
CodeMantis is Fair Source — a step between open source and closed source. It is licensed under the Fair Core License (FCL-1.0-ALv2): the core is free to use for any purpose that does not compete with CodeMantis, and every version automatically converts to the open source Apache 2.0 license two years after its release. Versions released before the FCL adoption remain under their original MIT License. Learn more at fcl.dev and fair.io.
OpenCode is CodeMantis's third first-class coding agent, wrapped through the Agent Client Protocol (ACP). Claude Code and Codex are tied to a vendor subscription (Anthropic and OpenAI); OpenCode isn't — it runs open and local models through providers you configure: OpenRouter (bring an API key), a local Ollama or LM Studio endpoint (no key, fully offline), or the free OpenCode Zen gateway. You pick it per session just like the others (it carries an ACP badge), and it plugs into the same interface — tool approvals, session history, SpecWriter, per-task routing, and Self-Drive all work identically. Reach for OpenCode when you want open or local models, want to keep work off a metered vendor pool, or want to run a model that Claude and Codex don't offer.
On June 15, 2026, Anthropic restructured Claude subscription billing — Agent SDK and `claude -p` (programmatic) usage moved to a separate, metered credit pool, while interactive use of Claude Code stays on the original subscription. CodeMantis runs the user's own claude binary, which keeps us in the explicitly-permitted lane, but the change made the case for vendor optionality clear. We had already shipped an AgentAdapter abstraction internally, so adding Codex took a single release — and OpenCode, which brings open and local models over ACP, followed the same way. Per-task routing in v1.5.0 lets you split workloads across all three — write specs with Claude, run hot main chat on Codex, keep experiments on an open model via OpenCode, or whatever balance fits your pools. Read the full reasoning in our blog post: 'Why CodeMantis Now Supports Two Coding Agents (and Soon More).'
Built by a creator.
Powered by Claude Code, Codex & OpenCode.
Made for creators.
AI coding enables a new dawn of creators. Ideas that were impossible due to budget, time, or team constraints are suddenly within reach. CodeMantis was built entirely by a non-developer using Claude Code itself — from the Rust backend to the React frontend — as proof that the path from idea to product is now open to everyone.
When Anthropic restructured Claude subscription billing on June 15, 2026, we learned a hard lesson: the agent you build on can change the rules underneath you. So we added OpenAI Codex as a first-class second agent, then OpenCodefor open and local models — and the architecture to add more whenever they show up. The app you choose shouldn't lock you into one vendor's billing decisions.
CodeMantis exists to support the creator's journey. Not for the engineers who already know how to code, but for the innovators who finally can — on whichever coding agent makes sense for them today.
Free. Fair source. Built for Mac & Windows.
The CodeMantis core is free, licensed under the Fair Core License (FCL). No API keys, no telemetry. Just download, connect your Claude Pro/Max or ChatGPT Plus/Pro/Business subscription — or point OpenCodeat an open or local model — and start building. Three first-class coding agents, in the interface they've been missing.