OpenSearch GenAI SDK
OTel-native tracing and scoring for LLM applications. Instrument your AI workflows with standard OpenTelemetry spans and submit evaluation scores, all routed to OpenSearch through a single OTLP pipeline.
Features
- One-line setup:
register()configures the full OTel pipeline (TracerProvider, exporter, auto-instrumentation) observe(): one decorator and context manager that creates OTel spans with GenAI semantic convention attributesenrich(): add model, token usage and other GenAI attributes to the active span from anywhere in your codemark_agent_root(): mark the agent's span as the trace's agent root in OpenSearch Agent Traces- Auto-instrumentation: discovers and activates installed instrumentor packages (OpenAI, Anthropic, Bedrock, LangChain and more)
- Scoring:
score()emits each evaluation result as an OTelgen_ai.evaluation.resultlog event, parented to the evaluated span, at span, trace, session or tool-call level - Benchmarks:
evaluate()runs your agent against a dataset with scorers;Benchmarkuploads results from any eval framework (RAGAS, DeepEval, pytest). Each case is its own agent trace plus onegen_ai.evaluation.resultlog event per score - AWS SigV4: built-in SigV4 signing for AWS-hosted OpenSearch and Data Prepper endpoints
- Zero lock-in: remove a decorator and your code still works; everything is standard OTel
Requirements
- Python: 3.10, 3.11, 3.12 or 3.13
- OpenTelemetry SDK: >=1.38.0, <2 (Logs API with
event_name, used for evaluation events)
Installation
pip install opensearch-genai-observability-sdk-py
Upgrading from 0.2.x? endpoint= is now the OTLP base URL and the default moved to http://localhost:4318: see Upgrading to 0.3.0.
The core package includes the OTel SDK and exporters. Auto-instrumentation of LLM libraries is opt-in: install the extras for the providers you use, for example pip install "opensearch-genai-observability-sdk-py[openai,anthropic]".
| Extra | Installs |
|---|---|
openai, anthropic, bedrock, google, langchain, llamaindex |
The instrumentor for that provider or framework |
otel-instrumentors |
All instrumentors, including Cohere, Mistral, CrewAI, MCP and vector stores |
opensearch |
opensearch-py, for OpenSearchTraceRetriever |
all |
Everything, for development and integration testing |
The full list is in Auto-instrumentation.
Quick start
Tracing
from opensearch_genai_observability_sdk_py import Op, enrich, observe, register
register(service_name="research-app") # OTLP to http://localhost:4318
@observe(name="research_agent", op=Op.INVOKE_AGENT)
def research(query: str) -> str:
with observe("web_search", op=Op.EXECUTE_TOOL):
results = [f"Result for: {query}"]
enrich(model="gpt-4.1", provider="openai", input_tokens=150, output_tokens=50)
return f"Summary of: {results}"
research("What is OpenSearch?")
This produces the span invoke_agent research_agent with a child execute_tool web_search. A longer example and the span tree are in Architecture.
Evaluations
This continues the tracing example: it scores research, and the evaluation events need the LoggerProvider that register() installs.
from opensearch_genai_observability_sdk_py import evaluate, scorer
@scorer(name="accuracy", threshold=1)
def accuracy(input, output, expected):
return 1.0 if expected in output else 0.0
result = evaluate(
name="research-agent",
task=research,
data=[{"input": "What is OpenSearch?", "expected": "OpenSearch"}],
scores=[accuracy],
)
Each case is its own agent trace, and each score is one gen_ai.evaluation.result event. To score live traffic, call score(): see score() and Getting started with agent evaluations.
Documentation
| Document | What it covers |
|---|---|
| docs/api-reference.md | Every public API: register(), observe(), Op, enrich(), mark_agent_root(), score(), evaluate(), Benchmark, OpenSearchTraceRetriever |
| docs/configuration.md | Environment variables, endpoints, protocols, AWS SigV4, ingestion pipeline settings, auto-instrumented libraries |
| docs/architecture.md | How spans and evaluation events flow from the SDK into OpenSearch |
| docs/agent-evaluations.md | Evaluation concepts and reference: data model, what gets stored, index template, ingestion, verify, troubleshooting |
| docs/getting-started-evaluations.md | Guided walkthrough from an instrumented agent to online and offline results in OpenSearch Dashboards |
| examples/ | Runnable examples, one SDK feature each |
The docs index is docs/README.md.
TypeScript SDK
Looking for the TypeScript version? See opensearch-genai-observability-sdk-ts.
Contributing
See CONTRIBUTING.md for the development setup, tests and the pull request process.
Security
See SECURITY.md to report a vulnerability.
License
This project is licensed under the Apache License 2.0.
Copyright
Copyright OpenSearch Contributors. See NOTICE for details.
Metadata
Release files for opensearch-genai-observability-sdk-py 0.3.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| opensearch_genai_observability_sdk_py-0.3.0.tar.gz | 113.4 kB | With provenance | Details for opensearch_genai_observability_sdk_py-0.3.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| opensearch_genai_observability_sdk_py-0.3.0-py3-none-any.whl | Python 3 | none | any | Details for opensearch_genai_observability_sdk_py-0.3.0-py3-none-any.whl |
Total release size: 179.5 kB
Release files / opensearch_genai_observability_sdk_py-0.3.0.tar.gz
| Download URL | opensearch_genai_observability_sdk_py-0.3.0.tar.gz |
|---|---|
| Size | 113.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
4811207c495aaf90005d6c9a5296f9720f47250544193b1b2b40f23fb077306d
|
|
BLAKE2b-256 checksum How to use checksums |
7e10a7870def062631e9a823351cf27a9b30328561b06bb247b0d579a42f2c37
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 18, 2026.
Transparency logRelease files / opensearch_genai_observability_sdk_py-0.3.0-py3-none-any.whl
| Download URL | opensearch_genai_observability_sdk_py-0.3.0-py3-none-any.whl |
|---|---|
| Size | 66.1 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
517a445e7f795a563edc3bd25827170a25e0222b2bb3712704701207cd742494
|
|
BLAKE2b-256 checksum How to use checksums |
1b37cebf3bfeaba5163b5b4702f4e10e252c527df7cb7f31a4b582da8678583c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.16
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 18, 2026.
Transparency logSecurity for opensearch-genai-observability-sdk-py 0.3.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for opensearch_genai_observability_sdk_py-0.3.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Nov 18, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.