PROTOCOL

Nexus A2A.
Secure Communication Between Healthcare AI Agents.

The messaging protocol that lets clinical AI agents delegate tasks to each other safely. Built on JSON-RPC 2.0 (JavaScript Object Notation Remote Procedure Call) with 13-point route admission validation on every request.

Book a Walkthrough API Reference
25
Clinical Agents
7,000+
Test Scenarios
13
Admission Checks
68
Personas
HOW IT WORKS

Task lifecycle in four steps.

Every agent-to-agent interaction follows a deterministic lifecycle. The requesting agent never loses visibility, and every state transition is logged to a hash-chained audit trail.

Step 01

Request

An agent publishes a task request specifying capability needed, patient context, urgency, and purpose of use. The request is signed with the sender's Decentralised Identifier (DID).

→
Step 02

Accept

The receiving agent passes all 13 admission checks. If every gate clears, it accepts the task and acknowledges with a session token bound to both agent identities.

→
Step 03

Checkpoint

During execution, the working agent emits checkpoint events via Server-Sent Events (SSE). The requesting agent maintains real-time visibility into progress and intermediate results.

→
Step 04

Complete / Escalate

The task completes with a structured result, or escalates to a human clinician if confidence thresholds are not met. Both outcomes are recorded in the audit trail.

JSON-RPC 2.0SSEWebSocket DIDmTLSFHIR R4
ROUTE ADMISSION

13-point validation on every call.

Before any agent can accept a delegated task, the request must pass all 13 admission checks. A single failure rejects the route and logs the reason. No exceptions outside emergency override.

1
Record Not Revoked — the agent's registration in the Global Agent Registry has not been revoked or suspended
2
Zone Match — both agents operate within the same logical security zone or an approved cross-zone bridge exists
3
Protocol Compatibility — the target agent supports the requested Nexus protocol version and message format
4
Endpoint Reachable — the target agent's endpoint responds to a health probe within the configured timeout
5
Trust Anchor Valid — the Decentralised Identifier (DID) document resolves and its cryptographic keys have not expired
6
Tag Policy — the requested task tags align with the receiving agent's declared capability tags and scope limitations
7
Jurisdiction — the patient data involved does not cross a jurisdictional boundary that either agent is prohibited from bridging
8
Version — the semantic version of the agent's capability contract is compatible with the caller's expectations
9
Capability — the target agent declares the specific capability required to fulfil the task
10
Authentication — mutual Transport Layer Security (mTLS) handshake succeeds and both agent certificates are valid
11
Consent — the patient's consent record authorises the specific data sharing required for this task
12
Purpose of Use — the declared purpose (treatment, payment, operations, research) is permitted by both agents' policies
13
Emergency Override — if invoked, validates that the request qualifies as a clinical emergency per the configured escalation criteria
REFERENCE IMPLEMENTATION

HelixCare: 25 agents across 5 clinical groups.

HelixCare is the reference implementation that demonstrates Nexus in a realistic clinical environment. 25 agents organised into 5 clinical demo groups, each handling a distinct healthcare workflow.

Command Centre

Central dashboard providing real-time visibility into all agent activity, task queues, admission decisions, and escalation events across the network.

On-Demand Gateway

API (Application Programming Interface) gateway that routes external requests to the correct clinical agent group, handling load balancing, rate limiting, and session affinity.

Triage & Assessment

Agents that perform initial patient assessment, symptom classification, urgency scoring, and clinical pathway selection before delegating to specialists.

Care Planning

Agents that generate and manage longitudinal care plans, coordinate multi-disciplinary team input, and track outcome measures over time.

Diagnostics & Imaging

Agents that coordinate lab orders, interpret diagnostic results using AI models from providers such as OpenAI, and flag abnormal findings for clinical review.

Prescribing & Pharmacy

Agents managing medication reconciliation, interaction checking, formulary compliance, and electronic prescribing workflows.

CLINICIAN AVATAR

Lip-synced 3D consultation interface.

A 3D avatar interface that delivers clinician-patient consultations using established medical communication frameworks. Every persona follows structured clinical methodology, not free-form conversation.

Calgary-Cambridge Framework

Consultations follow the Calgary-Cambridge guide to the medical interview — initiating the session, gathering information, building the relationship, explanation and planning, and closing the session. Every stage is tracked.

SOCRATES Pain Assessment

For pain presentations, the avatar uses the SOCRATES framework: Site, Onset, Character, Radiation, Associations, Time course, Exacerbating and relieving factors, and Severity.

ABCDE Emergency Protocol

Emergency presentations follow the ABCDE approach: Airway, Breathing, Circulation, Disability, and Exposure — ensuring systematic assessment and prioritisation of life-threatening conditions.

68 Personas, 3 Countries

The avatar system supports 68 distinct clinician personas spanning 3 countries. Each persona carries jurisdiction-specific clinical knowledge, language preferences, and regulatory awareness.

IDENTITY & TRUST

Every agent has a verifiable identity.

Nexus agents are not anonymous services. Every agent holds a Decentralised Identifier (DID), belongs to an Identity and Access Management (IAM) group, and is registered in the Global Agent Registry (GHARRA).

DID-Based Identity

Each agent's identity is anchored to a Decentralised Identifier (DID) document containing its public keys, service endpoints, and capability declarations. Identity verification happens on every request, not just at registration.

IAM Groups & Policies

68 personas are organised into Identity and Access Management (IAM) groups with fine-grained policies. Groups define which agent capabilities a persona can invoke, which patient data scopes are permitted, and which jurisdictions apply.

Global Agent Registry

Every agent is registered in GHARRA (Global Healthcare Agent Registry and Routing Architecture). The registry provides discovery, health monitoring, capability indexing, and revocation checking for the entire agent network.

Audit & Non-Repudiation

Every task request, admission decision, checkpoint, and completion is recorded in a hash-chained audit trail. Cryptographic signatures ensure non-repudiation — no agent can deny an action it performed.

See the protocol in action

Cross-Border Orchestration Demo Agent Network Topology Demo
Book a 30-Minute Architecture Walkthrough

Governed agent communication.

Nexus provides guaranteed delivery for clinical and AI workflows. 13-point admission control on every interaction. Explore the protocol specification or book a walkthrough to see it in action.

Book a Walkthrough API Reference

JSON-RPC 2.0 • 7,000+ test scenarios • 13-point admission control