Core concepts

A few ideas run through every Augsidius model. Understanding them makes every endpoint easier to use.

Clinical concepts and concept ids

Everything the API reasons about is a clinical concept in the Augsidius Evidence Graph: a condition, a complaint, an examination finding, a measurement, a historical finding, an investigation, a drug, a procedure, a therapy, an anatomical site, a demographic or lifestyle factor, and so on.

Each concept has a stable concept id (for example c9f1d2…) and a preferred name. Responses always include both, so you can store the id and show the name. You can send either free text ("neck stiffness") or a concept id; text is matched to concepts using names and synonyms.

Concept ids are stable within an API version. If two concepts are merged, the old id keeps resolving to the new one.

Observations

An observation is a fact about the patient expressed as a concept and a state:

Ruling something out is as informative as confirming it. Always send negatives you know about ("no rash", "not pregnant").

Care models

A care model is an adaptive clinical pathway built around observations. It is not a fixed flowchart. Every time an observation is added, the care model:

1. re-evaluates the whole observation set against the evidence graph;
2. works out which hypotheses are more or less likely;
3. identifies the missing observations that matter most — ranking first those that rule out emergencies and severe outcomes;
4. maps those missing observations to actions (ask, examine, measure, test) and ranks actions by how much they would clarify the picture;
5. (optionally) filters actions to what your care team can actually do — the provider scope;
6. remembers the next best actions it proposed, so the loop continues as results come in.

See Care models.

Cards

Decision-support models return cards — small, typed units of advice designed to be shown in a clinical workflow:

Cards use the HL7 CDS Hooks 2.0 card format, so EMR components that render CDS Hooks cards can show them unchanged. See Decision cards for every card kind and field.

Certainty labels

Every evidence-backed statement carries a certainty label on a GRADE-aligned scale:

Labels describe certainty of the evidence, not the strength of a recommendation.

Audience

Most models accept audience:

• clinician (default) — clinical language, full detail;
• patient — plain language, no dosing, no diagnostic labels presented as conclusions, always with safety-netting advice;
• machine — minimal prose; rely on the typed payload.

Patient audience requires the patient-facing terms to be accepted for the project in Graphite.

Units

Usage is measured in units, not tokens. A simple, rules-based call costs 1 unit; richer reasoning costs more (see Rate limits & usage). Your plan includes a monthly allowance of units for production keys; sandbox use is free within its limits.

Environments