Two independent profiles
Cybreach Engine keeps two AI profiles, configured separately: the project assistant on one side, the auto-audit engine on the other. They can point at different models: a fast one for drafting, a more capable one for parsing tool output.
Ollama, locally
This is the option to prefer when your reports contain sensitive client data: nothing leaves your infrastructure.
- Enter the endpoint of your Ollama instance
- Use model listing to retrieve what is available locally
- Adjust temperature and maximum tokens to suit the model
OpenAI-compatible endpoint
Any service exposing an OpenAI-compatible API will do. Provide the endpoint, the model name and the API key.
Your prompts and your project context (vulnerabilities, infrastructure, attachments) are sent to whichever provider you configure. Check that this transfer is compatible with your client confidentiality commitments.
Test before saving
The test button sends a real prompt and returns latency, the model response and any error, without saving anything. Use it systematically: an invalid key or a mistyped model name would otherwise only surface on first real use.