Overview
Trix seamlessly integrates with major AI model providers, allowing you to use memories with your preferred language model. Trix handles the embedding and retrieval, while you maintain control over your AI conversations.
Supported Providers
- OpenAI — GPT-4 class models and embeddings
- Anthropic — Claude family (Opus, Sonnet, Haiku)
- Google — Gemini models
- Mistral — Mistral / Mixtral models
- xAI — Grok models
- Custom Models — self-hosted or other API-compatible models
For Trix's own AI features (chat, enrichment, answer synthesis) you don't have to pin a specific model — request a capability tier and let Trix route to the best-fitting model, factoring in price, latency, and live provider health. See Model tiers & routing.
How It Works
Trix integrates at the prompt level:
- Your application queries Trix for relevant memories
- Trix returns semantically similar memories
- You inject these memories into your prompt context
- The AI model uses this context to generate responses
- Optionally, store new memories from the conversation
This approach keeps you in control of your AI pipeline while adding persistent memory capabilities.
Embedding Models
Trix uses state-of-the-art embedding models for semantic search:
- Default - Trix's optimized embedding model
- OpenAI - Compatible embedding models (optional)
- Custom - Bring your own embeddings (Enterprise)
Contact support to configure custom embedding models for your organization.