Skip to content

Releases

OHDSI 2026 Global Symposium

Choose a tag to compare

OHDSI 2026 Global Symposium

This release records the ACP R-client component exercised in the successful OHDSI 2026 Global Symposium Study Agent demonstration.

  • Commit: 8d02e87cbd78b04e25c11f58588a873beff3838e
  • Coordinated provenance: AtlasWebAPISandbox manifest
  • This is a conference provenance release. Technical workflow validation does not establish clinical validity or production readiness.

slashOhdsiAcpClient v0.0.3

Choose a tag to compare

@rkboyce rkboyce released this 10 Sep 17:42
5be702e

Highlights

  • Extends typed ACP client support for review-gated phenotype make-computable workflows.
  • Supports scope confirmation, concept-set review artifacts, source-informed conversion, and technical validation handoffs used by the Strategus shells.
  • Improves client integration for contextual workflow guidance and cohort-definition authoring flows.

Notes

Current supported modes include for the ACP call to phenotype_make_computable triggered by the "create=new definition" option of the "Source for [target/comparator/outcome]" step:

◦ direct Drug entry;
◦ positive supporting Condition occurrence in a window;
◦ Drug exposure followed by a Condition, with a clean window covering that same drug and condition.

They do not support “Drug A exposure with no prior exposure to Drug Class B.” Choosing “supporting Condition” for SGLT2 is semantically backwards: it would require a condition occurrence, not exclude prior SGLT2 drug exposure.

This release supports technical workflow integration; clinical concept and phenotype decisions remain explicitly user-reviewed.

v.0.0.2

Choose a tag to compare

@rkboyce rkboyce released this 05 Sep 16:15

Highlights

  • Supports the ACP client integration used by the review-gated narrative-to-computable phenotype and Keeper workflow enhancements.
  • Includes documentation and client-surface updates used by the Strategus R workflow shells.

Runtime provenance

Manual validation used muse-glimmer:latest, model digest de878ce33ad8 (recorded 2026-09-05): muse-glimmer architecture, 27.9B model parameters, Q4_K_M quantization; CLIP projector, 1.92B parameters, Q4_K_M quantization. The serving stack was Open WebUI with Ollama, distributed over two NVIDIA A30 GPUs. Model defaults were temperature=1, top_k=64, and top_p=0.95.

muse-glimmer:latest is a mutable label; the digest above is the release-time provenance record.

Compatibility

This project is under active development. Backward compatibility is not guaranteed at this time.

v0.0.1

Choose a tag to compare

@rkboyce rkboyce released this 20 Aug 13:05
adding R user-facing documentation