Which models Mesrai uses
The recommended catalog covers the major providers. You can bring any compatible model via BYOK:
- OpenAI GPT family (GPT-5.4 + derivatives)
- Anthropic Claude family (Sonnet 4.6, Opus 4.7 + derivatives)
- Google Gemini family (Gemini 3.1 Pro + derivatives)
- OpenAI-compatible providers — Moonshot, Z.ai, Fireworks, Together, Novita, OpenRouter, self-hosted Ollama / vLLM
How your code is handled
Code is sent to the LLM only for the duration of the review, then discarded. It is never used to train any model:
- Held only while a review is in flight
- Dropped from memory the moment the review completes
- Not used for training, fine-tuning, or evaluation
- Privacy guarantees follow the agreement you have with each LLM provider
Provider data-handling terms
Each provider publishes its own terms. Pick the link that matches the model you’re using:
- OpenAI: Enterprise privacy
- Google Gemini: Gemini API terms
- Anthropic Claude: How Anthropic uses personal data
What Mesrai does on its side
- Encryption — TLS in transit, AES-encrypted at rest
- Isolation — review sandboxes are ephemeral, destroyed after each run
- Access control — workspace roles + RBAC (see Workspace Roles)
- Regular audits — security review cadence + dependency scanning
- Patch discipline — security updates ship quickly, never sit in backlog
Questions
Email info@mesrai.com.