Sovraine/gliner-pii-onnx
Token Classification • Updated • 33
Sovraine
company
AI & ML interests
None defined yet.
Recent Activity
Organization Card
Sovraine
AI agent security — local, fail-closed, audited.
Sovraine builds Sovraine Guard, the governance layer for AI agents: a local proxy that intercepts every MCP tool call your AI clients make (Claude Desktop, Cursor, VS Code, Windsurf, ...), scans it for prompt injection, secrets, and PII, and enforces policy — entirely on-device, with zero cloud dependency.
The models published here are the on-device scanning engines that ship inside Sovraine Guard. They are ONNX exports optimized for local inference: no PyTorch, no GPU required, no data leaving your machine.
Models
| Model | Task | Used in Guard for |
|---|---|---|
| prompt-injection-onnx | Text classification | Detecting prompt-injection attempts in agent tool traffic |
| gliner-pii-onnx | Token classification (zero-shot NER) | Detecting PII/PHI in 11 languages for the privacy router |
Links
- Product & docs — Sovraine Guard repository and documentation
- open-guard-hub — community taxonomy, policies, and MCP tool mappings (CC-BY-SA-4.0)
- sovraine.com — company site
SOVRAINE PTE. LTD. — Singapore
datasets 0
None public yet