API reference¶
Generated from the docstrings of the opengrade-ai package (sdk/src/opengrade). For a guided tour with examples, read the Python SDK guide first.
Everything documented here is importable from the top-level package, whatever module it is defined in:
from opengrade import Opengrade
og = Opengrade(project="my-app") # tracking_uri defaults to http://localhost:8000
@og.trace
def answer(question: str) -> str: ...
result = og.evaluate(dataset="gold.jsonl", task=lambda item: answer(item.input), metrics=[...])
| Page | Contents |
|---|---|
| Tracing | init_tracing, flush, @trace, span, span attributes, users / sessions / tags |
| Experiments | evaluate, start_experiment, Experiment, ExperimentResult, sweep, evaluate_retrieval, chat test cases (chat_input, set_search_query) |
| Metrics and judges | metric, LocalMetric, Sample, Score, LlmJudge |
| Production scores and feedback | score_trace, log_score, log_feedback |
| Control plane client | Client, AsyncClient, AppConfigClient |
| Configuration and errors | Opengrade, Settings, active project, exceptions |