TealClassifier

pip install tealtiger

import asyncio
from tealtiger.classifier import TealClassifier

classifier = TealClassifier(
    mode="ensemble_union",
    model={
        "path": "./models/content-classifier-v2.onnx",
        "tokenizer": "./models/tokenizer.json",
    },
    confidence_threshold=0.7,
    categories=["pii", "secrets", "harmful", "off_topic"],
)

async def main():
    await classifier.warmup()

    result = await classifier.classify(
        "Please send payment to [email protected], account 4111-1111-1111-1111"
    )

    print(result.flagged)       # True
    print(result.highest_risk)  # "pii"
    for cat in result.categories:
        print(f"  {cat.name}: {cat.confidence:.2f} (source={cat.source})")

asyncio.run(main())

class TealClassifier:
    def __init__(
        self,
        mode: str = "ensemble_union",
        model: dict | None = None,
        confidence_threshold: float = 0.7,
        categories: list[str] | None = None,
        custom_patterns: dict[str, list[str]] | None = None,
        fallback: dict | None = None,
    ) -> None: ...

    async def classify(
        self,
        content: str,
        ctx: dict | None = None,
    ) -> ClassificationResult: ...

    async def classify_batch(
        self,
        contents: list[str],
    ) -> list[ClassificationResult]: ...

    def get_model_info(self) -> ModelInfo: ...

    async def warmup(self) -> None: ...

from dataclasses import dataclass

@dataclass
class ClassificationResult:
    categories: list[CategoryDetection]  # Detected categories
    flagged: bool                         # Any category above threshold
    highest_risk: str | None             # Highest-risk category
    duration_ms: float                   # Classification duration
    mode: str                            # Mode used
    sources: list[str]                   # "regex", "ml", or both

@dataclass
class CategoryDetection:
    name: str           # Category name
    confidence: float   # 0.0 – 1.0
    source: str         # "regex" | "ml" | "ensemble"
    details: list[str]  # Specific sub-patterns detected

classifier = TealClassifier(
    mode="ensemble_union",
    model={
        "path": "./models/classifier.onnx",
        "tokenizer": "./models/tokenizer.json",
        "max_length": 512,
        "provider": "cpu",  # "cpu" | "cuda" | "tensorrt"
    },
    confidence_threshold=0.7,
    categories=["pii", "secrets", "harmful"],
    custom_patterns={
        "internal_ids": [r"ACME-\d{6}", r"EMP-[A-Z]{2}\d{4}"],
        "medical": [r"\b\d{3}-\d{2}-\d{4}\b", r"MRN:\s*\d{8}"],
    },
    fallback={
        "mode": "regex_only",  # Fall back on ML failure
        "log_fallback": True,
    },
)

# Regex detects email pattern, ML confidence is 0.5 (below threshold)
# Result: FLAGGED (regex alone is sufficient in union mode)

# Regex misses novel PII format, ML confidence is 0.85
# Result: FLAGGED (ML alone is sufficient in union mode)

# Regex detects email pattern, ML confidence is 0.5 (below threshold)
# Result: NOT FLAGGED (both must agree)

# Regex detects email pattern, ML confidence is 0.85
# Result: FLAGGED (both agree)

from tealtiger import TealEngine, TealClassifier

engine = TealEngine(
    policy={
        "classifier": {
            "mode": "ensemble_union",
            "confidence_threshold": 0.7,
            "categories": ["pii", "secrets", "harmful"],
            "actions": {
                "pii": "auto_sanitize",
                "secrets": "auto_deny",
                "harmful": "auto_deny",
            },
        }
    },
    modules=[
        TealClassifier(
            mode="ensemble_union",
            model={"path": "./models/classifier.onnx", "tokenizer": "./models/tokenizer.json"},
            confidence_threshold=0.7,
            categories=["pii", "secrets", "harmful"],
        )
    ],
)