evalbase 0.0.2

Evalbase SDK for AI workflow telemetry using OpenTelemetry

Evalbase

Evalbase is a lightweight, easy-to-use Python SDK for instrumenting AI workflows with OpenTelemetry.

Installation

To install Evalbase from PyPI, run:

pip install evalbase

To install from source, clone this repository and run:

git clone https://github.com/evalbase/evalbase-python-sdk.git
cd evalbase-python-sdk
pip install .

Usage

1. Configuration: You can configure telemetry for your AI application by calling configure_telemetry.

from evalbase import configure_telemetry

configure_telemetry(
    service_name="my-ai-service",
    endpoint="https://otel-http.staging.evalbase.ai",  # Optional override
    api_key="apikey-123"  # Or set EVALBASE_API_KEY in your environment
)

2. Workflow and Step Decorators: Mark your functions with @workflow or @step(...) to automatically collect tracing and metrics.

from evalbase import workflow, step, StepType

@workflow
def my_workflow():
    result = my_llm_step("example")
    return result

@step(type=StepType.LLM)
def my_llm_step(prompt):
    # Your LLM logic here
    return "some LLM response"

3. Example: For a complete example of how to implement and test a workflow with steps, see the integration_tests directory.

python -m integration_tests.test_live_telemetry

For more information, check out the source code in decorators.py, telemetry.py, and integration_tests/test_live_telemetry.py.