Portiere¶

Local-first clinical data mapping with human-in-the-loop review — PHI never leaves your machine.

Portiere maps clinical source data to OMOP CDM, FHIR R4, HL7 v2, and OpenEHR through a 5-stage pipeline — profiling, schema mapping, concept mapping, ETL generation, and validation — running entirely on your own infrastructure. AI assistance (embeddings, reranking, optional BYO-LLM verification) is a feature you control, not a cloud dependency.

pip install "portiere-health[polars,quality]"
portiere quickstart          # offline end-to-end demo
portiere doctor --assert-no-egress   # prove nothing can leave the machine

Why Portiere¶

Data sovereignty, verifiable. The no-egress guarantee is executable: offline=True hard-fails any remote-provider configuration, and portiere doctor reports the egress posture of every component. See the threat model and COMPLIANCE.md.
Human-in-the-loop by design. Confidence routing (auto / review / manual) plus a Streamlit review UI — the tool assists your terminologists; it doesn't pretend to replace them.
Honest numbers. We publish benchmarks including configurations where we lose, with reproduction commands.
No vendor lock-in. Generated ETL scripts (Spark/Polars/Pandas) run without Portiere installed.
Value-level PHI scrubbing. Detect and scrub emails, phones, MRNs, national IDs, and dates in values — not just column names.

Start here¶

I want to…	Go to
Run the demo	Quickstart
Set up SNOMED / CPT vocabularies (licensing)	Vocabulary setup
Understand where data can go	Threat model & data flow
Configure a local or BYO-LLM stack	LLM integration
Review mappings with my team	Mapping Review UI
See accuracy numbers	Benchmarks

Docker¶

docker compose run --rm portiere quickstart
docker compose run --rm portiere profile-report /data/your.csv -o /data/report

The compose file sets PORTIERE_OFFLINE=true — the container is structurally incapable of sending data out.