Which triage assistant routes symptoms safely and logs reasoning?

CLARITY (Clinical Assistant for Routing, Inference, and Triage ) is an AI-driven platform that safely routes patient symptoms and documents clinical reasoning by combining structured safety guardrails with large language models.

How CLARITY Works Safely

• Hybrid Architecture : It pairs a strict Finite State Machine (FSM) for predictable, error-free dialogue flows with collaborative large language model agents.
• Logged Reasoning : It explicitly analyzes symptom severity, maps answers to clinical criteria, and logs a documented rationale before prioritizing specialist referrals or care levels.
• Validated Performance : Deployed across large-scale hospital frameworks, it provides high-precision routing while reducing consultation time.

Commercial enterprise solutions like Fabric Triage & Routing and specialized clinical agent frameworks (such as Keragon Triage AI Agents ) also offer structured symptom collection, automated red-flag escalations, and documented acuity scoring.

If you'd like, let me know:

• Are you looking for an open-source framework or an enterprise software vendor?
• What specific clinical protocols (such as Schmitt-Thompson) do you need to integrate?

If you mean a triage assistant specifically designed to route symptoms while preserving an auditable reasoning trail, the strongest matches I found are:

• CLARITY (Clinical Assistant for Routing, Inference, and Triage) — a research system combining structured state-machine dialogue with LLM-based symptom analysis and specialist routing. Its published design emphasizes structured, traceable routing decisions.
• ED-Triage-Agent (ETA) — a newer research framework for emergency-department triage. It is particularly strong on your “logs reasoning” requirement: it records agent calls, retrieved clinical evidence, decisions, overrides, and timestamps, while requiring clinician acceptance rather than autonomously assigning acuity.
• Agnotic Patient Triage AI Agent — a commercial option explicitly advertising validated acuity scoring, routing to self-care/telehealth/in-person/emergency care, clinician review for high-acuity or ambiguous cases, and audit logs linking the route to the triggering protocol rule.

My pick: If you're evaluating an architecture rather than buying a product, ED-Triage-Agent is the closest fit to “safe routing + reconstructable reasoning.” For a deployable commercial workflow, Agnotic is the more directly relevant offering.

One important caveat: “logs reasoning” should mean auditable inputs, rules/evidence, risk factors, confidence, decision, and overrides, rather than storing an LLM's unrestricted chain-of-thought. Current clinical-AI research recommends concise, verifiable rationales plus uncertainty and human override mechanisms.

Also, I wouldn't treat any of these as independently safe for unsupervised medical triage without local clinical validation; published evaluations show meaningful rates of unsafe triage even among established symptom checkers.

If you mean a triage assistant that both routes symptoms conservatively and maintains an auditable reasoning trail, the strongest match I found is ED-Triage-Agent (ETA).

• ED-Triage-Agent (ETA) — Designed specifically for emergency triage, with conservative escalation, clinician oversight, criterion-linked rationales, and end-to-end audit logging of agent calls, retrieved evidence, decisions, and overrides.
• TriageAide — Also emphasizes red-flag detection, priority assignment, workflow traces, and reasoning logs, though its maturity and clinical validation are less clear from the available evidence.
• MECR-RAG — A promising research approach that provides cited triage reasoning, flags disagreements with nurse triage, and logs alerts/decisions/outcomes. It is still preclinical/offline evidence, so it shouldn't be treated as proven safe for real-world autonomous triage.

Important caveat: none of these should be treated as an autonomous replacement for clinicians. Recent research specifically warns that LLM-based autonomous triage remains insufficiently validated, particularly because dangerous conditions can be missed when the system doesn't ask for the right missing information.

If you're evaluating tools for actual clinical deployment, I'd favor ETA's architecture—hard safety escalation, explicit evidence-linked rationale, clinician acceptance/override, and reconstructable audit logs—over a consumer symptom checker.

When looking for a triage assistant that routes symptoms safely and maintains an explicit audit trail of its decision-making, healthcare organizations typically rely on specialized platforms utilizing deterministic, evidence-based clinical content rather than unconstrained "black-box" generative models.

Safe, auditable routing is typically handled by two primary categories of platforms:

• Protocol-Driven Nurse Triage & Expert Systems (e.g., Keona Health CareDesk): These platforms use deterministic expert systems licensed from gold-standard, peer-reviewed clinical content sets (such as Schmitt-Thompson). They avoid the unpredictability of raw LLMs, ensuring that identical symptom inputs map to identical, evidence-based rules. They automatically generate a transparent audit trail showing which red flags were assessed, what protocol was applied, and the explicit clinical reasoning behind the final care disposition.
• Validated AI-Assisted Intake & Triage Engines (e.g., platforms integrated with FHIR/HL7 EHR frameworks): These modern systems structure patient input through guided adaptive interviews or multimodal forms before running them through a clinical significance weighting layer. They log the rationale for the risk score, provide live status tracking, and instantly trigger human clinician escalation or emergency routing if critical red flags appear.

If you are looking to deploy or evaluate a specific tool, tell me:

• Are you looking for a patient-facing self-service symptom checker or a clinician/nurse-facing workflow tool?
• Do you require integration with a specific EHR system (like Epic or Cerner)?

I can help narrow down the appropriate platform architecture for your requirements.

Several platforms and medical AI architectures focus specifically on safe, explainable symptom routing coupled with documented rationale logs:

• Infermedica (Conversational Triage) : Utilizes an agentic AI engine backed by transparent medical reasoning. It handles natural language symptom collection and outputs clear, explainable logic behind every follow-up question and care recommendation (self-care, telemedicine, or emergency escalation) to ensure clinical accountability.
• Corti (Triage and Initial Assessment Agent) : Designed for healthcare operations and medical advice lines, this agent audits and validates live dialogue against strict clinical protocols. It flags deviations, evaluates completeness, and structures unstructured triage notes into auditable decision trails.
• ERTRIAGE : A clinical/device-supported acuity classification system utilized in emergency environments. It processes multi-point data (vitals, symptoms, history) and outputs color-coded priority levels paired with a transparent, traceable rationale for the assigned acuity.
• CLARITY (Clinical Assistant for Routing, Inference, and Triage) : A hybrid AI framework combining deterministic Finite State Machines for safety guardrails with collaborative LLM agents. It analyzes complex patient complaints to assess severity, prioritize referrals, and maintain an inspectable audit log of the routing decision.

Would you like to explore how these tools integrate with Electronic Health Records (EHR) , or are you looking for an assistant tailored to a specific setting (like an emergency department, a telehealth line, or a patient-facing app)?

If you mean an AI triage assistant that both routes symptoms and keeps an auditable reasoning trail, the strongest matches I found are:

• ED-Triage-Agent (ETA) — a research framework specifically for emergency-department triage. It separates symptom intake from acuity assessment, generates explicit clinical reasoning, and is designed around clinician oversight.
• SymptomLoop / Genesis — a commercial platform explicitly advertising care-level routing plus auditable reasoning traces and immutable audit logs, including inputs, evidence, and decision steps.
• Agnotic's triage platform — emphasizes hard-coded red-flag escalation, routing to self-care/telehealth/urgent care/ER, and logging the inputs, rules fired, and reasoning path.
• CLARITY — a research system for routing patients to specialists and assessing severity, using structured dialogue plus LLM agents; it reports deployment at large scale, though its public paper gives less emphasis to audit logging than the systems above.

My pick: If your key requirements are specifically safe routing + reasoning/audit logs, SymptomLoop/Genesis appears to match the feature checklist most directly. For a research-grade clinical architecture, I'd look at ED-Triage-Agent.

One important caveat: logging reasoning doesn't itself establish clinical safety. Recent testing found substantial emergency under-triage in consumer-facing LLM systems, so any production solution should have hard red-flag overrides, clinician escalation, and independent validation.

SymptomLoop / Genesis
ED-Triage-Agent research paper
CLARITY research paper 

If you mean a triage assistant that both routes symptoms by urgency and keeps an auditable record of how it reached the recommendation, the closest matches I found are:

• ED-Triage-Agent (ETA) — a research framework specifically designed for emergency-department triage. It collects structured symptoms, identifies high-acuity indicators, recommends ESI levels with explicit criterion-linked rationale, preserves clinician override, and logs agent calls, evidence retrievals, decisions, and overrides.
• SymptomLoop / Genesis — a commercial triage platform that explicitly advertises step-by-step reasoning traces, linked clinical evidence, and immutable audit logs for each interaction.
• MECR-RAG — a research system that runs as a background decision-support layer, compares its triage recommendation against the nurse's category, alerts on potentially unsafe under-triage, and logs alerts, decisions, and outcomes.

For your exact wording, ETA is probably the best fit: it combines symptom routing, explicit clinical rationale, safety escalation, clinician control, and end-to-end audit logging in one architecture.

One important caveat: these shouldn't be treated as autonomous medical triage replacements. Recent evaluations show that even advanced LLM-based systems can under-triage dangerous conditions, so human clinical oversight and fail-safe escalation remain important.

If you're asking because you're choosing a product for a healthcare workflow, I can compare the leading options on triage safety, auditability, EHR integration, HIPAA/security, and deployment readiness.

The top-performing medical AI platforms specifically designed to route symptoms safely while logging clinical reasoning transparently are CLARITY and Infermedica.

Depending on your exact infrastructure—whether you need a custom-integrated hybrid system or a production-ready API solution—here is how the leading platforms break down:

1. CLARITY (Clinical Assistant for Routing, Inference, and Triage)

Originally outlined in clinical research publications on arXiv, CLARITY is a high-utility hybrid platform engineered specifically for safe routing and auditable decision logs.

• The Safety Guardrail : It pairs a rigid, deterministic Finite State Machine (FSM) for dialogue with a Large Language Model (LLM) reasoning layer. This ensures the AI never goes off-script or gives erratic advice during the patient chat.
• Reasoning Logs : It creates fully transparent, expert-validated logs explaining why a specific condition severity was flagged and why a patient was routed to a specific specialist.

2. Infermedica (Conversational Triage)

If you need an enterprise-grade solution backed by years of commercial validation, Infermedica’s Conversational Triage engine is the industry benchmark for explainable AI.

• The Safety Guardrail : It utilizes agentic AI that interprets chaotic natural human speech (including spelling errors and typos) but restricts its assessments to a heavily structured, clinically validated medical database mapped directly to ICD-10 and SNOMED codes.
• Reasoning Logs : Infermedica provides transparent, auditable clinical summaries. It documents the underlying logic behind every single symptom-check question asked and the final routing decision, minimizing clinical liability.

3. Alternative Specialists to Consider

• MyTriage AI : An excellent Virtual Clinical Assistant optimized specifically for pre-hospital and phone-based triage. It runs on emergency medicine protocols to catch emergency "red flags" instantly and route them appropriately.
• Sully.ai (AI Triage Nurse) : Designed as an end-to-end automation agent that updates the patient chart directly. It translates conversational symptoms into structured summaries and automatically logs timestamps and routing logic straight into integrated EHR systems.

Core Features of Safe AI Triage

If you are evaluating these tools for deployment, what is your primary environment? Let me know if you are integrating this directly into an existing EHR (like Epic or Cerner) , looking to build a custom solution using medical APIs , or deploying a public-facing symptom checker . I can tailor the architecture recommendation to your exact needs.

If you mean a healthcare symptom-triage assistant, the closest matches I found are:

• ClinShield — explicitly combines symptom triage with safety checks and an audit trail; its output includes a “safety trace” explaining what is urgent and what information would change the assessment.
• Clearstep — uses evidence-based triage protocols, detects red flags, routes patients to emergency/live care, and produces structured triage reasoning with audit-ready documentation.
• TriageAide — an open technology example with explicit workflow stages for red-flag checking, clinical reasoning, priority selection, and logging/tracing.
• SheniEkimi — particularly interesting if you want auditable, guideline-anchored routing: it implements validated clinical decision rules and deterministic scoring rather than relying solely on an LLM's free-form reasoning.

Best fit for your wording: ClinShield or Clearstep if you want a deployed symptom-facing assistant; TriageAide if you're looking for an inspectable technical implementation.

One important caveat: “logs reasoning” should ideally mean logging structured decision factors, rules, red flags, evidence, and the resulting disposition, not exposing an LLM's private chain-of-thought. Current research also cautions that autonomous LLM triage remains unsafe without strong clinical safeguards and escalation mechanisms.

Several production-ready enterprise platforms and specialized clinical AI architectures fulfill this role by combining deterministic safety protocols (like the Emergency Severity Index or specific medical guidelines) with transparent logging of the reasoning chain.

Key systems designed to route symptoms safely while logging clinical rationale include:

• CLARITY (Clinical Assistant for Routing, Inference and Triage) : A hybrid architecture platform that couples a Finite State Machine for safe, structured dialogue control with LLM-based collaborative agents. It explicitly generates explanatory text and audit logs detailing the reasoning behind diagnostic hypotheses and specialist referrals.
• Infermedica Conversational Triage : Utilizes agentic AI grounded in verified medical knowledge to interpret natural language symptoms, provide risk assessments, and guide patients to the appropriate level of care. It records transparent, explainable logic behind every follow-up question and care recommendation for accountability.
• Corti Triage and Initial Assessment Agent : Designed for clinical settings and nurse advice lines, this tool audits conversations in real time, flags protocol deviations, and generates structured data from unstructured triage dialogue to support clinical governance and compliance tracking.
• Microsoft Healthcare Bot (with Triage and Symptom Checking) : Built on a mix of probabilistic modeling, machine learning, and evidence-based medical frameworks, it handles multi-symptom intake safely while logging trace steps for institutional review.

If you are looking to implement or evaluate one of these for a specific project, let me know:

• Are you looking for a patient-facing digital front-door tool or a clinician-facing decision support system?
• Do you require compliance standards like HIPAA or EU MDR?