Skip to content

Testing & Verification Guide

This page explains how to run automated unit and integration tests, as well as execute manual verification scripts to check system integrity, agent loops, and observability tracing.

🧪 1. Automated Testing (Pytest)

We use pytest (with pytest-asyncio) to run isolated configuration, loader, and integration tests.

Running the Tests

To run the automated test suite, activate your virtual environment and execute pytest, ignoring the manual tests folder: 

source venv/bin/activate
pytest --ignore=tests/manual

Automated Test Suite Descriptions

  • tests/test_config.py: Verifies that environment variable profiles (default, test, prod) reload variables (like OLLAMA_URL and MONGODB_URI) correctly from their respective .env files.
  • tests/test_tool_loader.py: Validates the standard tools dynamic dynamic initializer, verifying the execution of Python code definitions inside toolsConfig.json.
  • tests/test_integration.py: Asserts successful connectivity with a running Ollama container and verifies that the Agent is created with the expected endpoints.

🛠️ 2. Manual Testing Scripts (tests/manual/)

Manual test scripts are divided into language folders to separate browser automation tests from diagnostic script tests.

A. Python Verification Scripts (tests/manual/python/)

These scripts test specific API elements or simulate trace outputs directly inside the agent environment.

  • test_logging.py: Simulates two separate agent executions (one successful, one catching a tool division-by-zero boundary error) to verify that trace payloads and console logs are formatted correctly.
  • generate_trace.py: Runs a simple query through the ReAct agent, flushes the Langfuse callback client, and confirms trace logs.
  • generate_trace_simple.py: Directly invokes the Langfuse SDK to create a raw trace and span, verifying authorization keys.
  • verify_api.py / verify_api2.py: Connect to the local Langfuse API server endpoints via requests to fetch and print recent trace logs programmatically.
  • e2e_test.py: Launches the local Chainlit app in a background subprocess, uses Playwright Chromium to navigate to the web page, asserts successful UI render, and cleanly terminates the server.

B. Node.js Browser Scripts (tests/manual/js/)

These use Puppeteer to simulate human browser interaction, click UI elements, submit inputs, and scrape outputs to verify E2E tracing.

  • debug_chainlit.js: Launches a headless browser, navigates to http://localhost:8000, types a query, dispatches a keydown Enter event, waits for agent response, and captures screenshots to tests/manual/screenshots/.
  • verify_puppeteer.js / verify_puppeteer_fix.js: Automate input submission in Chainlit, then log into the Langfuse dashboard at http://localhost:3000 to verify that the trace for the submitted message exists in the database.
  • verify_traces.js: Inspects the Langfuse UI trace log table and prints the 5 most recent traces.