AI Agents

MEDrecord operates a fleet of 20+ specialized AI agents that automate healthcare workflows, improve documentation, and assist staff across all products.

Overview

AI Agents are purpose-built assistants that integrate with MEDrecord products to provide intelligent automation while maintaining full compliance and audit trails.

Key Capabilities

• Task Automation - Automate repetitive workflows
• Content Generation - Draft messages, documentation, and reports
• Data Analysis - Extract insights from healthcare data
• Decision Support - Provide recommendations (with human approval)
• Quality Assurance - Review content for compliance and accuracy

Agent Catalog

Communication Agents

Documentation Agents

Operations Agents

Analytics Agents

Security Agents

Architecture

Agents are accessed through the Gateway:

Product → Gateway → Agent Router → AI Provider
                         ↓
                    Agent Config
                    (prompts, context)

Agent Configuration

Each agent has:

• System Prompt - Role and behavior definition
• Context - Product-specific context injection
• Guardrails - Output validation rules
• Audit Level - Logging requirements

Integration

Calling Agents

const response = await fetch('/api/agents/message-drafter', {
  method: 'POST',
  headers: {
    'Content-Type': 'application/json',
    'X-Gateway-Signature': signature
  },
  body: JSON.stringify({
    context: { patientId, appointmentType },
    prompt: 'Draft appointment reminder'
  })
})

Human-in-the-Loop

All agent outputs require human approval before:

• Sending to patients
• Modifying records
• Changing configurations

This ensures compliance and maintains accountability.

Security and Compliance

• No PHI in Prompts - Patient data is referenced by ID, not content
• Audit Logging - All agent interactions are logged
• Output Validation - Responses are validated against schemas
• Rate Limiting - Per-agent and per-tenant limits

Creating Custom Agents

To create a custom agent:

1. Define the agent purpose and scope
2. Create system prompt with guardrails
3. Configure context injection
4. Set up audit logging
5. Test with sample inputs
6. Deploy via Gateway configuration

See Developer Tools for AI development best practices.

Resources