Feature Request: Assign Multiple AI APIs to Roo Code Modes

[nader443]

🧩 Summary

Roo Code currently supports multiple working modes (e.g., Code Mode, Architect Mode, Ask Mode, etc.), but is limited to using a single AI provider at a time. I propose enabling a multi-AI routing feature that allows users to assign different AI APIs to each mode — unlocking better cost control, performance optimization, and task specialization.

---

🧠 Rationale

Different AI models are optimized for different use cases — and Roo Code’s modular mode-based architecture is an ideal fit for assigning mode-specific AI engines.

Below is an example mapping of Roo Code modes to AI models that suit their respective purposes:

Roo Code Mode	Example AI Model	Rationale / Ideal Fit
Code Mode	claude-3.5-haiku	Fast, efficient, cost-effective for general-purpose coding tasks
Architect Mode	claude-3.7-sonnet	Strong reasoning and verbose output for design & planning
Ask Mode	gemini-pro	Versatile, conversational — ideal for quick Q&A
Debug Mode	gpt-4o or mistral-large	Good at tracing problems, explaining stack traces, suggesting fixes
Orchestrator Mode	claude-3-opus	Strategic, capable of managing subtasks and reasoning across them

📝 Note: These are example mappings — users should be able to customize model selection based on their access, budget, and preferences.

Key benefits:

• 🔄 Performance Optimization — match the right AI engine to each workflow
• 💸 Cost Efficiency — use cheaper models for frequent, low-complexity tasks
• ⚙️ Customizable Experience — user control over engine assignments
• 🧪 Experimentation — test new APIs per mode without global impact

---

⚙️ Suggested Implementation

Add a new section in Roo Code's settings file:

// ~/.roo/settings.json
{
  "aiEngines": {
    "codeMode": "claude-3.5-haiku",
    "architectMode": "claude-3.7-sonnet",
    "askMode": "gemini-pro",
    "debugMode": "gpt-4o",
    "orchestratorMode": "claude-3-opus"
  }
}

Then update internal logic to route requests by mode:

function getAIEngine(mode: string): string {
  return settings.aiEngines[mode] ?? settings.defaultAI;
}

---

📦 Optional Enhancements

• ✅ Fallback model per mode (for quota errors or unavailability)
• 🔁 Weighted ensemble support (advanced: mix outputs from multiple models)
• 🌐 Plugin architecture for community-contributed adapters

---

🙏 Closing Thoughts

This feature would greatly extend Roo Code’s flexibility and unlock better user control. The mode-based architecture is already a perfect match for this enhancement — it just needs AI routing logic baked in.

I’d be happy to contribute to design, test, or help spec this out further!

Thanks for your time — looking forward to your thoughts.