88% of Agent projects fail because they lack evaluation infrastructure. agenttest gives you familiar test syntax to assert Agent behavior — intents, entities, sentiment, PII, hallucinations, and more.
import { agentTest, expect } from '@agenttest/core' agentTest('customer service agent handles refund', async (agent) => { const response = await agent.send('I want a refund for order #1234') expect(response).toMentionIntent('refund') expect(response).toContainEntity('orderNumber', '1234') expect(response).toHaveSentiment('empathetic') expect(response).not.toContainPII() expect(response).toRespondWithin(5000) // 5 seconds expect(response).not.toHallucinate() // grounded in context })
Verify your agent recognizes the right intent from user messages — refund, tracking, complaint, escalation.
Assert that your agent correctly identifies entities like order numbers, emails, and customer names.
Ensure your agent responds with the right tone — empathetic, positive, neutral — matching the situation.
Catch leaked SSNs, credit cards, and phone numbers before they reach your users.
Detect when your agent invents facts not grounded in the provided context.
Deterministic LLM mocks for unit testing — fast, repeatable, no API calls needed.
This Pages deployment is documentation and a product preview. It does not accept test runs, JWT credentials, SQLite data, Stripe payments, or API keys. Run the SDK locally with bun test; connect the existing Docker dashboard through Cloudflare Tunnel before enabling accounts.