Multi-AI Assistants
Use the latest models from OpenAI, Google, and Anthropic alongside your main coding session — works with Claude Code or Codex.
Overview
CodeMantis lets you create additional AI assistant tabs that run alongside your main coding session. These assistants use the chat APIs from OpenAI, Google, and Anthropic, giving you access to the latest models from each provider without leaving the app.
The Assistant panel is agent-agnostic — it works the same whether your main session is powered by Claude Code or Codex. The assistants are chat-only — they cannot edit files or run commands in your project. They are designed for tasks where you need a second opinion, a different model's perspective, or a parallel conversation that doesn't interfere with your coding session.
You can also use local CLI agents (Claude Code or Codex itself) as assistants — see the multi-CLI selector when creating a new assistant tab.
Setting Up API Keys
Before using assistants, add your API keys in Settings > API Keys:
- OpenAI — For the latest GPT models
- Google — For Gemini models
- Anthropic — For Claude models via the API (separate from your Claude Code subscription)
Keys are stored securely in the macOS Keychain. You only need to enter them once. For a detailed walkthrough of setting up and testing API keys, see Setting Up AI API Keys.
Creating Assistant Tabs
To create a new assistant:
- Click the + button in the tab bar.
- Select New Assistant.
- Choose a provider (OpenAI, Google, or Anthropic).
- Pick a model from the available options.
Each assistant opens in its own tab. You can have multiple assistants open simultaneously, each using a different provider or model.
Provider Badges
Every assistant tab displays a colored provider badge so you can quickly identify which model you're talking to. This prevents confusion when you have several tabs open at once.
Use Cases
Assistants are useful for tasks that complement your main coding workflow:
- Brainstorming — Explore architectural approaches before committing to an implementation.
- Code Review — Paste code into an assistant tab and ask for a review without interrupting your main session.
- Documentation — Draft README files, API docs, or comments in a separate tab.
- Comparison — Ask the same question to multiple models and compare their answers.
- Research — Look up framework APIs, language features, or best practices.
Per-task routing. In v1.5.0, you can route the Assistant panel itself to a specific agent in Settings › Agents › Per-task Defaults (the
assistantrow). See Per-Task Agent Routing.
Cost Tracking
CodeMantis tracks token usage and estimated cost for each assistant session. You can view this in the assistant tab header, which shows the running token count and approximate spend. This helps you monitor API costs across providers.
Limitations
Assistants operate in a chat-only mode. They cannot:
- Read or write files in your project
- Execute terminal commands
- Access MCP servers
- Interact with your main coding session
If you need the AI to make changes to your codebase, use the main session (Claude Code or Codex).