What is Verial?
Verial is a simulation and evaluation platform for healthcare AI agents. It lets you spin up realistic healthcare environments (EHRs, payer portals, clearinghouses, phone systems, fax lines) and run structured benchmarks against them. Your agent interacts with simulated interfaces exactly as it would in production, and Verial evaluates the results using LLM-powered assertions.- Simulated healthcare interfaces. FHIR R4 servers, HL7v2 endpoints, fax lines, voice/IVR systems, payer portals, clearinghouses, CDS Hooks, and messaging.
- Structured benchmarks. Define tasks with patient scenarios, expected outcomes, and natural language evaluation criteria.
- Automated evaluation. An LLM judge evaluates agent behavior against your assertions, producing weighted scores and pass/fail verdicts.
- Full interaction capture. Every FHIR request, HL7 message, fax document, and phone call is recorded as evidence for evaluation.
Quick Start
Create an environment, run a benchmark, and check results in 5 minutes.
API Reference
REST API for managing environments, benchmarks, and runs.
How It Works
The core workflow:- Create simulators and an environment. Create simulator definitions (FHIR, voice, fax, etc.), then link them into an environment that models the health system your agent will interact with.
- Define a benchmark with tasks and evals. Each task describes a scenario (e.g., “Schedule a follow-up appointment for patient John Smith”) and includes evals that check whether the agent performed correctly.
- Run the benchmark. Verial provisions live sandbox instances from your simulators, your agent connects and performs the tasks, and all interactions are recorded.
- Review results. Each task receives a score based on its evals. The run produces an aggregate score and a pass/fail verdict.
Core Concepts
| Concept | Description |
|---|---|
| Environment | A composition of simulators representing a simulated health system |
| Simulator | A reusable definition of a simulated interface (FHIR server, phone line, payer portal) |
| Benchmark | A collection of tasks linked to an environment, with evals for scoring |
| Run | A single execution of a benchmark, producing scores and a verdict |
| Eval | A natural language assertion (label + assert) evaluated by an LLM judge |
Who Is Verial For?
Verial is built for teams developing healthcare AI agents. If your agent needs to interact with EHRs, submit prior authorizations, process faxes, handle phone calls, or work with clinical decision support systems, Verial gives you a safe, repeatable way to test those interactions before going live. Common use cases:- Prior authorization agents that submit requests to payer portals and track approvals
- Clinical documentation agents that read and write to EHR systems via FHIR
- Scheduling agents that handle phone calls and coordinate appointments
- Claims processing agents that work with clearinghouses and payer systems
- Care coordination agents that send faxes, messages, and manage referrals
SDK
TypeScript SDK for programmatic access to all Verial resources.
MCP Server
Connect AI agents directly to Verial via Model Context Protocol.