FerroEHR
The openEHR CDR validates the coded text of a composition against its terminology bindings through a FHIR terminology server.
FerroTERM serves the HL7 FHIR terminology API on R4, R4B, R5 and the R6 ballot from one binary over an index built once per release and read at startup. No JVM, no search cluster. You bring the release you are licensed for; the build tool turns it into an index; the server answers from it.
A server image and a console image on GHCR and the
ferroterm and ferroterm-build binaries for
Linux, built at SLSA Build Level 3, and the release readers and the
terminology engine on crates.io.
Every code system reaches the FHIR operations through the same provider seam, so nothing in an operation is a special case for one system.
$lookup, $validate-code,
$subsumes, $expand,
$translate and $closure, answered from
the index the server read at startup.
/r4, /r4b, /r5 and
/r6, in JSON or XML, each version's parameter set
generated from HL7's own packages.
Value sets by composition, the implicit value sets of SNOMED CT, LOINC, RxNorm and ICD-11, and the SNOMED CT Expression Constraint Language over the inferred hierarchy.
Code systems
SNOMED CT, LOINC, ICD-10 and ICD-11, RxNorm, the Dutch national
sets and any FHIR CodeSystem, each built from its own
release, with UCUM and the registries inside the binary.
Code systems, value sets and concept maps written to the server and kept with their history, when you name a database for them.
National terminology servicesThe sync service reads a national terminology service's feed on a schedule, builds what you subscribed to and asks the server to reload, in one direction only.
SNOMED CT, LOINC, UCUM, BCP 47, BCP 13, ISO 3166-1, ICD-10 (WHO), ICD-10-NL, ICD-10-CM, ICD-11 MMS, ICD-11 ICF, ICD-11 Foundation, ATC/DDD, ICPC-2, RxNorm, the DHD Diagnosethesaurus and Verrichtingenthesaurus, the G-Standaard, the Nederlandse Labcodeset, the NHG ICPC-1 to SNOMED CT map, and any FHIR CodeSystem, ValueSet, and ConceptMap resources. The code systems page lists each with its URI, versions, build command, and licence.
Every figure below comes from a benchmark record committed to the repository; a check fails when the page and the records disagree. The benchmarks page has every operation, the cold and tail latencies, the method, and how to take a record yourself.
$lookup, warm p50 · 91 MB resident · 44 MB on disk · built in 1.51 s
ICD-10-NL96 µsa $lookup, warm p50 · 66 MB resident · 21 MB on disk · built in 880.52 ms
ICD-11 MMS138 µsa $lookup, warm p50 · 99 MB resident · 44 MB on disk
LOINC136 µsa $lookup, warm p50 · 440 MB resident · 310 MB on disk · built in 8.15 s
RxNorm (prescribable subset)388 µsa $lookup, warm p50 · 133 MB resident · 77 MB on disk · built in 4.58 s
SNOMED CT (International edition)217 µsa $lookup, warm p50 · 943 MB resident · 559 MB on disk · built in 14.12 s
SNOMED CT (Netherlands edition)222 µsa $lookup, warm p50 · 1.14 GB resident · 864 MB on disk · built in 17.66 s
Warm p50 over 200 HTTP round trips on one machine (Apple M2, 17.18 GB, macos/aarch64), FerroTERM 0.1.4 serving FHIR R4B, taken 2026-09-22. Every tile links to its record.
The console is its own program and image,
ferroterm-console, beside the server. It reaches the
server over the public FHIR API, so anything it does any FHIR
client can do, and it edits the resources your deployment writes.
The server and the build tool ship in one image and one tarball. The registry systems need nothing loaded; every other code system is an index you build once per release.
docker run of the published image serves UCUM,
BCP 47, BCP 13 and ISO 3166-1 with no configuration, and answers
a $lookup at once.
compose.yaml ships with every release. Point it at
the index directory you built; the console profile
adds the console.
ghcr.io/ferrohealth/ferroterm and
ghcr.io/ferrohealth/ferroterm-console, for amd64 and
arm64, each with SLSA provenance and an SBOM you verify before
you run it.
The FerroHEALTH products reach each other only over published protocols, and any FHIR client can use FerroTERM. These are the family's.
The openEHR CDR validates the coded text of a composition against its terminology bindings through a FHIR terminology server.
The bridge asks for code systems and value sets through
$lookup, $translate and
$validate-code while it maps openEHR to FHIR.
The patient index translates every HL7 v2 table value of its v2 face on a FHIR terminology server.
What it serves and how well, how it compares to the reference servers on the same machine, conformance per FHIR version, and the licence.
Every operation under each version's base, in JSON or XML, with worked examples.
Installing it, loading a code system, sizing the hardware, and verifying a release before you run it.
Building and testing it, the generated FHIR model, and the performance gate every change passes.