Quickstart¶

The shortest path to see piighost at work, without downloading a model. You will de-identify a sentence from a dictionary of known values, in under a minute.

import asyncio

from piighost.components.detector import ExactMatchDetector
from piighost.pipeline import AnonymizationPipeline

detector = ExactMatchDetector({"John Doe": "PERSON", "Paris": "LOCATION"})
pipeline = AnonymizationPipeline(detector)


async def main() -> None:
    result = await pipeline.anonymize("John Doe lives in Paris.")
    print(result.text)


asyncio.run(main())

The output should be:

<<PERSON:1>> lives in <<LOCATION:1>>.

How it works¶

ExactMatchDetector spots exact occurrences, at word boundaries, of the values in the dictionary you pass. Given only a detector, the pipeline fills the required stages with their defaults. It groups the detections of the same value and label with an ExactEntityLinker, then replaces each entity with a LabelCounterPlaceholderFactory that numbers per label, so <<PERSON:1>> and <<LOCATION:1>>. Overlap resolution runs by default, and the optional stages, entity expansion, entity resolution, override, and guard, stay disabled. That is enough for a first try, with no model to load.

What's next¶

• For real automatic detection, arbitrary names and locations, move on to the First pipeline with an NER like GLiNER2.
• To describe a full pipeline in a file rather than in Python, see the TOML reference.
• To de-identify across a conversation with persistent memory, see the Conversational pipeline.