🥋Sensei

The open-source qualification engine for AI agents

Test. Evaluate. Certify.
Before you hire an agent, ask the Sensei.

View on GitHubRead the Docs

Three-Layer Evaluation

Every agent is tested across three dimensions. No shortcuts.

Task Execution

Task Execution

Can the agent do the job?

Measure real performance against domain-specific KPIs. Each task is scored on concrete, quantifiable metrics — not vibes.

Reasoning

Reasoning

Can it explain its decisions?

Probe the agent’s thought process. Great execution means nothing if the agent can’t articulate why it made a choice.

Self-Improvement

Self-Improvement

Can it learn from feedback?

Give the agent feedback and watch it adapt. The best agents don’t just perform — they evolve.

See It In Action

Watch Sensei evaluate an agent in real-time. Pick a suite and see how the three layers unfold.

sensei — evaluation
▋

How It Works

A simple, structured pipeline from agent to verdict.

🤖Your AgentHTTP / Stdio
🥋SenseiEngine
🎯 Task
🧠 Reason
📈 Growth
📊Score0–100
🏅BadgeDecision

Built-In Test Suites

Battle-tested evaluation suites for the most common agent roles. Create your own in minutes.

📞12 scenarios

SDR

Cold outreach, email personalization, call analysis, and pipeline qualification

🎧15 scenarios

Support

Ticket resolution, multi-turn conversations, escalation handling, CSAT optimization

✍️10 scenarios

Content Writer

Blog posts, social copy, SEO optimization, brand voice consistency

🧪14 scenarios

QA Engineer

Test case generation, bug reporting, regression analysis, coverage assessment

📊11 scenarios

Data Analyst

SQL generation, insight extraction, visualization recommendations, anomaly detection

💻16 scenarios

Developer

Code generation, refactoring, PR review, documentation, debugging

Three Lines to Qualify

Load a suite, create an adapter, run. That's it.

qualify.ts
import { SuiteLoader, Runner, createAdapter } from '@sensei/engine';
const suite = await loader.loadFile('./suites/sdr-qualification/suite.yaml');
const adapter = createAdapter(suite.agent);
const result = await new Runner(adapter).run(suite);
// result.scores.overall → 87.3
// result.badge → "silver"