Same agent Smaller bill. The tools and infrastructure to get more work from your AI. Fewer tokens. Less compute. Lower cost.
TypeScript agent-harnessagent-harness-engineeringai-engineeringai-observabilityamd-gpuclaude-codeclickhousecodexcoding-agentcoding-agentscursorgenaigpu-monitoringllm-observabilityllmopsnvidia-smiobservabilityopen-sourceopenaiopentelemetry
# pylint: disable=missing-function-docstring
"""Provider-coverage matrix test for ``openlit.agent.version_hash``.
Every Python LLM provider instrumentation MUST stamp the agent version hash
on its chat span (see ``apply_agent_version_attributes`` in
``openlit.__helpers``). This lets the unified Agents page filter by
``openlit.agent.version_hash`` directly instead of falling back to the
materialized version time window, which is both slower and imprecise.
This test guards against regression: if someone adds a new chat-LLM
provider but forgets to wire ``apply_agent_version_attributes``, this test
fails. Coverage is asserted statically (reading the provider's ``utils.py``
once) so we do not have to mount full mock provider clients for each one.
For ``strands``, the equivalent stamping lives in
``instrumentation/strands/processor.py`` (a span processor, not a wrapper);
it is checked alongside ``utils.py`` for that provider.
"""
from __future__ import annotations
from pathlib import Path
import pytest
PROVIDERS_DIR = (
Path(__file__).resolve().parents[1] / "src" / "openlit" / "instrumentation"
)
# Providers that wrap a chat-LLM call and therefore MUST stamp the hash.
# Keeping the list explicit (rather than auto-discovering every utils.py)
# avoids false positives on RAG/agent/vector/etc. instrumentations that
# don't have a chat span to stamp.
LLM_PROVIDERS_WITH_UTILS = (
"ai21",
"anthropic",
"azure_ai_inference",
"bedrock",
"cohere",
"gpt4all",
"gradient",
"groq",
"litellm",
"mistral",
"ollama",
"openai",
"premai",
"pydo",
"reka",
"sarvam",
"together",
"transformers",
"vertexai",
"vllm",
)
# Providers that stamp via a span processor instead of a wrapper utils.py.
PROCESSOR_PROVIDERS = ("strands",)
@pytest.mark.parametrize("provider", LLM_PROVIDERS_WITH_UTILS)
def test_provider_utils_stamps_version_hash(provider: str) -> None:
utils_path = PROVIDERS_DIR / provider / "utils.py"
assert utils_path.is_file(), f"missing utils.py for provider '{provider}'"
source = utils_path.read_text(encoding="utf-8")
assert "apply_agent_version_attributes" in source, (
f"provider '{provider}' missing apply_agent_version_attributes; "
"every chat-LLM provider must stamp openlit.agent.version_hash "
"on its chat span"
)
@pytest.mark.parametrize("provider", PROCESSOR_PROVIDERS)
def test_processor_provider_stamps_version_hash(provider: str) -> None:
processor_path = PROVIDERS_DIR / provider / "processor.py"
utils_path = PROVIDERS_DIR / provider / "utils.py"
sources = []
if processor_path.is_file():
sources.append(processor_path.read_text(encoding="utf-8"))
if utils_path.is_file():
sources.append(utils_path.read_text(encoding="utf-8"))
assert sources, f"provider '{provider}' has neither processor.py nor utils.py"
combined = "\n".join(sources)
assert (
"apply_agent_version_attributes" in combined
or "OPENLIT_AGENT_VERSION_HASH" in combined
), (
f"provider '{provider}' (processor-based) must stamp "
"openlit.agent.version_hash on its chat span"
)