Hi Minerva team,
We've recently implemented an initial Minerva export backend in the SKaiNET project and wanted to share our progress as well as open a discussion for future collaboration.
https://github.com/SKaiNET-developers/SKaiNET
https://skainet-developers.github.io/SKaiNET/skainet/how-to/minerva-export.html
What we've implemented
- Minerva export pipeline inside SKaiNET
- Graph validation and compatibility checks
- Conversion of SKaiNET models into Minerva-compatible NPZ artifacts
- Automated invocation of the Minerva compiler toolchain
- Project packaging for firmware and host-side verification
- Optional host verification workflow
Our understanding of Minerva's security value
We identified Minerva's core security contributions as:
- Encrypted model weights
- Integrity verification
- Output authentication and replay protection
- Runtime tamper/fault detection mechanisms
- PRNG-based activation protection
- Deterministic embedded-friendly execution model
Feedback
The integration worked well overall and the architecture is clean and easy to automate from external tooling.
While integrating, we noticed a few areas where additional documentation could help:
- Long-term API stability expectations
- Recommended key provisioning workflow
- Clarification around current runtime APIs versus legacy examples
- Future roadmap for additional targets and model architectures
Thanks for building an interesting project. We'd be happy to share diagrams, findings, and integration feedback if useful.
Hi Minerva team,
We've recently implemented an initial Minerva export backend in the SKaiNET project and wanted to share our progress as well as open a discussion for future collaboration.
https://github.com/SKaiNET-developers/SKaiNET
https://skainet-developers.github.io/SKaiNET/skainet/how-to/minerva-export.html
What we've implemented
Our understanding of Minerva's security value
We identified Minerva's core security contributions as:
Feedback
The integration worked well overall and the architecture is clean and easy to automate from external tooling.
While integrating, we noticed a few areas where additional documentation could help:
Thanks for building an interesting project. We'd be happy to share diagrams, findings, and integration feedback if useful.