tailtester 0.1.0

The pytest for AI agents — auto-generate and run tests for any AI agent

tailtest

The pytest for AI agents.

"You don't write tests. You build your agent -- we watch, we learn, we test."

---

The Problem

93% of developers don't test their AI agents. The tooling doesn't exist, the patterns aren't established, and the only serious option -- Promptfoo -- just got acquired by OpenAI. There is no vendor-neutral, open-source, developer-first testing tool for AI agents. If you ship an agent today, you're shipping it blind.

What This Will Be

• Position 0: Observes your development process and auto-generates tests
• Deterministic + LLM-judged + red-team assertions in a single framework
• Any framework: LangChain, CrewAI, PydanticAI, OpenAI Agents SDK, raw API calls
• Any model: OpenAI, Anthropic, Google, Ollama, anything via litellm
• CLI-first, CI/CD native -- exit codes, JUnit XML, parallel execution
• Built-in red-teaming: prompt injection, jailbreak, PII extraction, OWASP compliance
• Production monitoring with automatic regression test generation from failures
• Zero telemetry, fully local, Apache 2.0 -- no data leaves your machine, ever

Quick Start (Future)

pip install tailtester
tailtest scan .
tailtest run

Three commands. No config files. No account creation. Meaningful test results in under 3 minutes.

Example Test

from tailtest import agent_test, expect

@agent_test
async def test_order_lookup():
    response = await agent.chat("What's the status of order #12345?")
    expect(response).to_call_tool("lookup_order")
    expect(response).tool_called_with("lookup_order", order_id="12345")
    expect(response).to_contain("order")
    expect(response).no_pii()
    expect(response).latency_under(3000)
    expect(response).cost_under(0.50)

@agent_test
async def test_response_quality():
    response = await agent.chat("Explain your return policy")
    expect(response).faithful_to(context="Returns accepted within 30 days...")
    expect(response).helpful()
    expect(response).tone("professional", "empathetic")

@agent_test(retries=10)
async def test_reliability():
    response = await agent.chat("What are your business hours?")
    expect(response).to_contain("9am")
    expect(response).pass_rate(0.95)

Deterministic assertions (cost, latency, tool calls, PII) run instantly at zero cost. LLM-judged assertions (faithfulness, tone, quality) default to a local model via Ollama.

Architecture

+-------------------+     +-------------------+     +-------------------+
|  CONTEXT ENGINE   | --> |  TEST GENERATOR   | --> |   TEST RUNNER     |
|                   |     |                   |     |                   |
|  Scan codebase    |     |  Deterministic    |     |  Parallel exec    |
|  Watch file edits |     |  LLM-judged       |     |  Record / replay  |
|  Ingest OTel      |     |  Red-team         |     |  CI/CD mode       |
|  Detect framework |     |  Regression       |     |  JUnit XML output |
+-------------------+     +-------------------+     +-------------------+
                                                            |
                                                            v
                                                    +-------------------+
                                                    |  ASSERTION ENGINE |
                                                    |                   |
                                                    |  Deterministic    |
                                                    |  LLM-judged       |
                                                    |  Reliability      |
                                                    +-------------------+

What We Are NOT Building

• Not a dashboard-first enterprise product (that's Braintrust)
• Not a framework-specific tool (that's LangSmith)
• Not a security-only scanner (that's Promptfoo/OpenAI now)
• Not a cloud-required service (runs fully local, forever)

Current Status

Phases 1-4 complete. The core engine is built and working. Pre-launch, not yet published.

Metric	Value
Python files	~135
Lines of code	~22,000
Internal tests	328 passing in 2.11s
CLI commands	13 (init, scan, run, generate, redteam, watch, guard, ingest, record, replay, report, doctor, mcp-serve)
Assertion types	22 (10 deterministic + 7 LLM-judge + 5 reliability)
Framework detectors	6 (OpenAI, Anthropic, LangChain, CrewAI, PydanticAI, generic)
Red-team attacks	64 across 8 categories
OWASP checks	20 (LLM Top 10 + Agent Top 10)
MCP server tools	6
Report formats	5 (terminal, JUnit XML, JSON, compliance text, compliance HTML)
Example projects	4 (hello-world, openai-assistant, crewai-research, raw-api-agent)
Commits on main	15

See examples/ for sample agent projects demonstrating the full pipeline.

Tech Stack

• Python 3.11+ with uv for package management
• Click for CLI, Pydantic v2 for data models
• litellm for model-agnostic LLM calls
• asyncio + httpx for parallel test execution
• opentelemetry-sdk for production trace ingestion

Contributing

This project is in early development. Contribution guidelines will be published soon.

License

Apache 2.0