Agents & governance

The gathering isn’t skipped — it’s done, and validated.

EMME does the data-gathering for you, but it doesn’t ask you to take it on faith. Each step is a separate agent with one job and clear limits, working over graded, provenance-linked evidence — and every output is a draft you review and sign.

How the evidence is validated

Nothing is autonomous

Every output is a draft. The clinician reviews and signs — and the sign-off records who, when, and the exact text approved. EMME is decision support; the clinician is the validating authority.

Evidence is graded, not asserted

MASE carries an evidence tier, a 0–100 confidence, provenance and dissent on every claim. Governed evidence is clearly distinguished from AI-synthesised (labelled “verify”). Studied ≠ proven; traditional use ≠ trial evidence.

The patient validates their own history

During intake the patient confirms the history back to EMME before it is finalised — so the record starts from the patient's own attestation, then the clinician reviews it.

A de-identification boundary

Only de-identified clinical questions ever reach the evidence engine. Patient identifiers never leave EMME for MASE.

MASE draws on 29.4 million biomedical research papers, graded for consensus and conflict.

Each agent, and what it will & won’t do

Intake & documentation

Gathers a full history conversationally and drafts the documentation.

What it does

Builds a structured Living Timeline; screens for red flags; drafts SOAP notes, referrals and letters.

Its limits

Does not diagnose, advise, or decide. The patient confirms accuracy; the clinician reviews.

Analysis agent

Maps patterns and proposes the questions worth researching.

What it does

Surfaces obstacles, symptom clusters and safety flags, and frames the evidence queries.

Its limits

Internal only — never shown to the patient. It frames what to investigate; it does not conclude.

MASE — Multi-Agent Synthesis Engine

Retrieves and grades the evidence.

What it does

Draws tier-scored, provenance-linked research from the biomedical corpus and grades consensus vs conflict.

Its limits

Governed evidence carries tier + confidence + source, with dissent intact. Where it's absent, output is labelled AI-synthesised — verify.

Protocol agent

Drafts the evidence-grounded care plan.

What it does

Grounds each recommendation in the MASE evidence (carrying its tier and dissent), with differentials and a consult agenda.

Its limits

No dosing, no diagnosis — you prescribe. Every section is yours to accept, edit, reject or workshop.

Safety agent

Screens every protocol for harm.

What it does

Checks recommended substances against the patient's medications, conditions, allergies and pregnancy status — worldwide, with severity tiers.

Its limits

Flags signed protocols for re-review when reference data changes; it advises, it does not override your judgment.

Patient Summary agent

Writes a plain-language note for the patient.

What it does

A warm, 7th-grade-reading-level summary, released to the patient only when you choose.

Its limits

No diagnosis, mechanisms, differentials, medication names or dosing.

Record extraction agent

Reads a patient-uploaded document and summarises it for the record.

What it does

Extracts a plain-language summary and structured facts (results, medications, conditions, dates) from an uploaded lab result, letter or discharge summary.

Its limits

Reports only what the document says — no diagnosis, interpretation or advice. Flags anything unreadable; the original is always attached.

The final check

You are the validation.

Export and patient-sharing are blocked until you sign. Sign-off captures who signed, when, and the exact text approved, over an append-only audit of the whole chain — data in → AI output → your review → sign-off → share. In a teaching setting, a student drafts and a supervisor signs.

See it on a real case.

Free trial — 5 workups, unlimited Scribe, no card.

Create your account