[`opentelemetry-instrumentation-genai-agno`] Add instrumentation for agno model calls by DylanRussell · Pull Request #773 · open-telemetry/opentelemetry-python-genai

DylanRussell · 2026-09-22T14:43:12Z

Description

Probably want to wait until #765 lands to avoid duplicate inference calls, but eventually we will want this and that PR will not change the implementation here.

I think once this lands we will have reached parity with the agno instrumentations in open inference and other places..

Type of change

New feature (non-breaking change which adds functionality)

How has this been tested?

Unot tests

Checklist

Followed the style guidelines of this project
Changelog updated if the change requires an entry
Unit tests added
Documentation updated

Instrument agno.models.base.Model response and streaming methods to emit standard GenAI chat spans with complete request/response attributes, usage metrics, and content capture. Assisted-by: Claude Opus 4.6

Replace Any with concrete Agno Model, Message, MessageData, and ModelResponse types in model patching, stream wrapping, and provider resolution functions. Assisted-by: Claude Opus 4.6

…tent variable Remove redundant capture_content local variables and function parameters, relying directly on invocation.should_capture_content property on InferenceInvocation and RetrievalInvocation. Assisted-by: Claude Opus 4.6

Replace repetitive try/except type conversions with safe_int and safe_float helpers across retrieval, model inference, and streaming wrappers. Assisted-by: Claude Opus 4.6

…hecks Let opentelemetry-util-genai handle None values directly across model response token metrics, embedder tokens, dimension count, and query text. Assisted-by: Claude Opus 4.6

Avoid repeated checks of getattr(model, 'vertexai', False) and GOOGLE_GENAI_USE_VERTEXAI by caching the boolean in is_vertex. Assisted-by: Claude Opus 4.6

Eliminate repetitive is_vertex if/else branching by determining google_provider once at function start. Assisted-by: Claude Opus 4.6

Consolidate duplicate provider resolution logic across resolve_embedder_provider and resolve_model_provider into a single _resolve_provider helper. Assisted-by: Claude Opus 4.6

…_model_output_message Drop unnecessary cast(object, ...) and cast(list[Any], ...) calls and share tool call part extraction via _extract_tool_call_parts. Assisted-by: Claude Opus 4.6

…nput_messages Use list[dict[str, Any]] | None for tool_calls and Iterable[Message] for messages, accessing Message attributes directly. Assisted-by: Claude Opus 4.6

…n embedder functions Use concrete Model | Embedder and Embedder annotations in _resolve_provider, resolve_embedder_provider, and embedder wrappers. Assisted-by: Claude Opus 4.6

…ment) Assisted-by: Claude Opus 4.6

… | ModelResponse Assisted-by: Claude Opus 4.6

Assisted-by: Claude Opus 4.6

opentelemetry-pr-dashboard · 2026-09-22T14:44:37Z

Pull request dashboard status

Waiting on reviewers · refreshed 2026-09-23 20:44 UTC

Review the latest changes.

Status above doesn't look right?

Just replied or pushed? Anything around or after the refresh time above may not be picked up yet — give it a few minutes.
Anything look wrong? Report it with what you expected; it helps us improve the dashboard.

Copilot

Copilot review overview

🟡 Changes recommended

Model spans start after inference, stream managers are wrapped incorrectly, and required tests are missing.

Get a fresh assessment by requesting another Copilot review.

Review effort: Balanced
Findings: 2 · 1 · 3

Open (6)

Wrap outer synchronous model call to capture latency and failures · New
Use synchronous stream manager wrapper for context-manager API · New
Decode JSON tool-call arguments before storing them · New
Add and register chat conformance scenario · New
Test synchronous stream early close and single finalization · New
Test async stream interruptions, errors, and early close · New

What changed in this PR

Adds Agno model-call telemetry for synchronous, asynchronous, streaming, multimodal, and error paths.

Changes:

Adds GenAI chat instrumentation and stream wrappers.
Adds provider/message/media extraction utilities.
Adds model tests and documentation.

File	Description
`patch.py`	Patches model response processing.
`stream.py`	Adds model stream telemetry wrappers.
`utils.py`	Adds model and multimodal extraction.
`test_model.py`	Tests model instrumentation.
`test_embedder.py`	Makes Gemini tests optional.
`README.rst`	Documents model operations.
`.changelog/773.added`	Records the feature.

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.