Using standard tests ↗

Designed to test the component with access to external services (in particular, the external service that the component is designed to interact with)

View API reference

Both types of tests are implemented as pytest class-based test suites.

Implementing standard tests#

Depending on your integration type, you will need to implement either or both unit and integration tests.

By subclassing the standard test suite for your integration type, you get the full collection of standard tests for that type. For a test run to be successful, the a given test should pass only if the model supports the capability being tested. Otherwise, the test should be skipped.

Because different integrations offer unique sets of features, most standard tests provided by LangChain are opt-in by default to prevent false positives. Consequently, you will need to override properties to indicate which features your integration supports - see the below example for an illustration.

# Indicate that a chat model supports image inputs

class TestChatParrotLinkStandard(ChatModelIntegrationTests):
    # ... other required properties

    @property
    def supports_image_inputs(self) -> bool:
        return True  # (The default is False)

You should organize tests in these subdirectories relative to the root of your package:

• tests/unit_tests for unit tests
• tests/integration_tests for integration tests

To see the complete list of configurable capabilities and their defaults, visit the API reference for standard tests.

Here are some example implementations of standard tests from popular integrations:

Unit tests Unit tests Unit tests

Integration tests Integration tests Integration tests

Ensure your integration passes the standard test suite. See the Daytona integration as an example.

Sandbox Integration tests

Sandbox integrations#

Deep agents sandbox integrations use SandboxIntegrationTests from langchain_tests.integration_tests. Subclass it and provide a sandbox fixture that yields a SandboxBackendProtocol instance. Use the Daytona integration tests as a reference implementation. See Contributing a sandbox integration for publishing guidelines.

Running tests#

If bootstrapping an integration from a template, a Makefile is provided that includes targets for running unit and integration tests:

make test
make integration_test

Otherwise, if you follow the recommended directory structure, you can run tests with:

# Run all tests
uv run --group test pytest tests/unit_tests/
uv run --group test --group test_integration pytest -n auto tests/integration_tests/

# For certain unit tests, you may need to set certain flags and environment variables:
TIKTOKEN_CACHE_DIR=tiktoken_cache uv run --group test pytest --disable-socket --allow-unix-socket tests/unit_tests/

# Run a specific test file
uv run --group test pytest tests/integration_tests/test_chat_models.py

# Run a specific test function in a file
uv run --group test pytest tests/integration_tests/test_chat_models.py::test_chat_completions

# Run a specific test function within a class
uv run --group test pytest tests/integration_tests/test_chat_models.py::TestChatParrotLinkIntegration::test_chat_completions

Troubleshooting#

For a full list of the standard test suites that are available, as well as information on which tests are included and how to troubleshoot common issues, see the Standard Tests API Reference.

Edit this page on GitHub or file an issue.

Connect these docs to Claude, VSCode, and more via MCP for real-time answers.