Multi-LLM: why Luna uses Claude, Gemini, and more

Depending on a single AI provider is a risk. Luna was designed from the start to be multi-LLM.

Why a single model isn't enough

Each LLM has its strengths and weaknesses:

• Claude (Anthropic) excels at structured reasoning and nuanced analysis
• Gemini (Google) has a massive context window and handles multimodal data well
• Groq/Mistral offer ultra-low latency for simple tasks

A single-model system also suffers from provider outages. When OpenAI goes down, so do you.

Luna's routing architecture

Our LLM router operates on three axes:

1. User preference

You choose in Settings > AI Engine your preference: Auto, Claude, or Gemini. Auto mode optimizes automatically.

2. Per-task routing

• Deduplication: fast model (Haiku/Flash)
• Cross-referencing: analytical model (Sonnet/Flash)
• Deep research (Mythos): advanced model (Opus) with extended thinking

3. Automatic failover

If Claude is unavailable, Luna switches to Gemini (and vice-versa) without service interruption. No missed alerts.

Bring Your Own Key

Luna goes further: you can use your own API keys for:

• Anthropic (Claude)
• Google (Gemini)
• Groq
• Mistral
• OpenRouter (access to 50+ models)

Your keys are stored encrypted and take priority over Luna's system keys. Result: your LLM costs are under your control.

Full transparency

Each scan indicates in the audit trail which model was used and why. No black box.