Jeanette/Docs

Core Concepts

AI Models & Reliability

How Jeanette uses multiple AI models across providers for reliability and performance.

Multi-model architecture

Jeanette doesn't rely on a single AI provider. Instead, she uses a pool of frontier AI models from multiple providers, so if one has issues, your experience isn't affected.

When you first start using Jeanette, you're assigned a primary AI model. This is your "home" model - the one Jeanette uses for your conversations by default. You'll stay on this model unless there's a problem, in which case Jeanette will automatically and transparently switch to another model until the issue resolves.

Frontier models

These are the high-quality models Jeanette uses for conversations, tasks, research, and email processing:

Model Provider
Claude Sonnet Anthropic
GPT-5.4 OpenAI
Gemini 2.5 Pro Google
Grok 3 xAI

All of these are top-tier, frontier-class models with strong reasoning, tool use, and natural language capabilities. You'll get a great experience regardless of which one you're assigned to.

Fast models

For quick background operations - like generating progress updates, summarising conversations, and extracting memories - Jeanette uses smaller, faster models:

Model Provider
Claude Haiku Anthropic
GPT-4.1 Mini OpenAI
Gemini 2.5 Flash Google

These handle the lightweight work so your conversations stay snappy without unnecessary cost.

Automatic failover

If your assigned model experiences issues (rate limits, outages, or errors), Jeanette doesn't just wait and retry the same provider. Instead, she immediately falls back to another healthy model in the pool:

  1. Your request is sent to your assigned model
  2. If it fails, Jeanette checks which other models are currently healthy
  3. She retries your request on the best available alternative
  4. You get your response without noticing anything happened

Once the original model recovers, Jeanette switches back automatically. Your model assignment doesn't change - the fallback is temporary.

How model health is tracked

Jeanette monitors the health of every model in real time. If a model fails three times within a minute, it's temporarily marked as unavailable for all users (since all requests share the same provider connection). After 10 minutes, Jeanette tests the model again and brings it back into rotation if it's recovered.

This means a provider outage affecting one user is detected immediately and prevents the same issue from hitting other users.

Will I notice a difference between models?

In most cases, no. Jeanette's personality, instructions, and behaviour are defined by her system prompt, not the underlying model. All frontier models handle tool calling, multi-step tasks, and natural conversation well.

That said, each model has subtle strengths - some are slightly more concise, others more detailed. If you ever feel like your experience has changed, it may be because of a temporary failover. It'll resolve on its own.