Vinci Cyber: open-weight cybersecurity models

AI for reviewable defensive security work

Given a configuration and a scanner finding, Vinci Cyber is trained to propose a bounded repair, leave a configuration unchanged when appropriate, or reconsider an unsuccessful attempt.

The intended audience is security teams and engineers investigating infrastructure configuration findings. It is a narrow defensive specialization, not a general vulnerability scanner or an autonomous remediation system.

SimpleDirect® specializes existing IBM Granite and Mistral Devstral foundations through post-training. These releases are not foundation models pretrained from scratch in Canada.

What cybersecurity evaluation establishes

Internal repair scoring for the 8B and 30B releases remains under qualification, with rescoring pending. The 123B detail page documents its separate evaluation record and limitations. Scanner clearance alone does not prove a valid security repair.

Use each artifact’s own evaluation record. Results for a parent model, sibling size or full-precision checkpoint do not establish the performance of a different release or quantized format.

Plan a self-hosted deployment

Public model cards provide weights and setup guidance for running Vinci Cyber on infrastructure you control. Independently check configuration syntax, security properties and regressions, then obtain human approval before applying a proposed change.

Allow memory beyond the weight files, choose a supported runtime, and record the exact revision and format. Tool use requires your own application, permissions and controlled workspace. Review and test suggested changes before applying them; these pages do not offer hosted inference.