{"entity":{"id":"local-llm-reasoning-layer","kind":"term","name":"Local-only language models over patient data (privacy by architecture)","aka":["local-only LLM","on-device LLM","on-device language model","local language model","egress guard","egress guards","cohort briefing","LLM reasoning layer","no-network inference"],"tldr":"Running an open language model on the same machine as a patient's data, with code that blocks any network call, lets a system explain model outputs in words without the data ever leaving the room.","summary":"A large language model is an AI model trained on vast text for generation and analysis (Wikipedia), and a clinical decision support system provides clinicians with knowledge and person-specific information in the workflow (Wikipedia). Cloud models cannot see identifiable patient data without agreements, so one design runs an open-weight model locally over a model's derived outputs only, with assertion-style guards that fail loudly if anything would be sent out or committed, and uses a cloud model only for cohort-level summaries of public findings. It is a privacy rule expressed as architecture, complementary to HIPAA and GDPR compliance.","asOf":"2026-09-24","wikipedia":"https://en.wikipedia.org/wiki/Large_language_model","links":[{"label":"Wikipedia: clinical decision support system","url":"https://en.wikipedia.org/wiki/Clinical_decision_support_system"},{"label":"Wikipedia","url":"https://en.wikipedia.org/wiki/Large_language_model"}],"tags":["cansim-terms"],"related":["cancer-ai-vocabulary"],"cancers":[],"sections":[],"technologies":[],"targets":[],"drugs":[],"companies":[],"institutions":[],"pathways":[],"terms":["hipaa","controlled-access-data","uncertainty-quantification","open-weights"],"trials":[],"people":[],"bottlenecks":[],"keyPapers":[],"journals":[],"dependsOn":[],"notes":["Listed in the CanSim terms map 1.0.0 (docs/onco/terms.json, generated 2026-09-24), CC BY 4.0, attribution: CanSim project, an open, public-data-first cancer foundation-model programme; CanSim page path /terms/local-only-patient-reasoning-layer."],"provenance":{"editedBy":"OnCo CanSim terms wave (Wikipedia summaries, standards and project pages, GDC and FDA pages, Europe PMC)","editedOn":"2026-09-24","note":"CanSim terms map 1.0.0 (docs/onco/terms.json, generated 2026-09-24), CC BY 4.0, attribution: CanSim project, an open, public-data-first cancer foundation-model programme"},"category":"Methods and models"},"route":"/terms/local-llm-reasoning-layer/","neighbours":{"term":[{"id":"cancer-ai-vocabulary","kind":"term","name":"Cancer AI vocabulary (CanSim terms map)","route":"/terms/cancer-ai-vocabulary/"},{"id":"controlled-access-data","kind":"term","name":"Controlled-access genomic data (dbGaP, EGA)","route":"/terms/controlled-access-data/"},{"id":"hipaa","kind":"term","name":"HIPAA (Health Insurance Portability and Accountability Act)","route":"/terms/hipaa/"},{"id":"open-weights","kind":"term","name":"Open weights, open code and gated models","route":"/terms/open-weights/"},{"id":"uncertainty-quantification","kind":"term","name":"Uncertainty quantification and confidence gates","route":"/terms/uncertainty-quantification/"}]}}