adding crew ai instrumentation

[venkata-pavani]

Description

Description

What I changed	What this means
Added CrewAI LLM call instrumentation	CrewAI LLM calls now produce telemetry that follows the OpenTelemetry GenAI semantic conventions.
Added GenAI attributes	The telemetry includes available model, provider, request settings, messages, tool definitions, response details, token usage, and error information.
Added spans and metrics	The instrumentation emits inference spans, client duration metrics, and token usage metrics through opentelemetry-util-genai.
Added success and failure handling	Successful calls complete their telemetry, while failed calls record and re-raise the original error without changing CrewAI's behavior.
Disabled CrewAI native telemetry by default	This prevents two unrelated sets of telemetry from being emitted while this instrumentation is active.
Allowed users to retain CrewAI native telemetry	Users can set CREWAI_DISABLE_TELEMETRY=false. Existing values are preserved and restored correctly. ref: #308 (comment)
Added completion-hook support	Captured prompts and completions can be forwarded using the shared completion-hook mechanism.
Added tests	Tests cover completed calls, failed calls, content capture, completion hooks, semantic-convention attributes, and native telemetry configuration.
Updated documentation and examples	The README explains the new behavior and includes a custom completion-hook example.

Type of change

Please delete options that are not relevant.

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

Future PR

In follow-up PRs, I plan to add CrewAI workflow, agent, and tool instrumentation using the invoke_workflow, invoke_agent, and execute_tool GenAI semantic conventions, along with proper context propagation and streaming support.

[opentelemetry-pr-dashboard]

Pull request dashboard status

Waiting on the author · refreshed 2026-08-09 04:43 UTC

Respond to 4 review items (e.g. link a commit, explain why not, ask a follow-up):

• Inline threads: 1, 2, 3, 4

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.
• Should this be with reviewers? Comment /dashboard route:reviewers to route it to them.
• Anything wrong — including the routing? Report it with what you expected; it helps us improve the dashboard.

[Copilot]

Pull request overview

Adds first-pass OpenTelemetry GenAI instrumentation for CrewAI LLM inference by translating CrewAI LLM lifecycle events into opentelemetry-util-genai inference telemetry, along with unit + conformance scaffolding and docs/examples.

Changes:

• Introduces a CrewAIInferenceEventListener that maps CrewAI LLM events to TelemetryHandler.inference() invocations.
• Adds unit tests for inference event translation, completion hook wiring, and disabling CrewAI’s native telemetry while instrumented.
• Adds a CrewAI conformance scenario + tox wiring and expands the package README with configuration and completion hook docs.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tox.ini	Adds a CrewAI conformance tox env and updates CrewAI test commands to exclude conformance from non-conformance runs.
instrumentation/opentelemetry-instrumentation-genai-crewai/tests/test_instrumentor.py	Adds tests ensuring CrewAI native telemetry is disabled only when instrumentation sets the default.
instrumentation/opentelemetry-instrumentation-genai-crewai/tests/test_inference.py	Adds unit tests validating span attributes/status for completed/failed events and completion hook invocation.
instrumentation/opentelemetry-instrumentation-genai-crewai/tests/test_conformance.py	Adds conformance test entrypoint wiring for CrewAI scenarios.
instrumentation/opentelemetry-instrumentation-genai-crewai/tests/conftest.py	Ensures content_capture="SPAN_ONLY" for tests that assert content-related attributes exist.
instrumentation/opentelemetry-instrumentation-genai-crewai/tests/conformance/inference.py	Adds a minimal inference conformance scenario emitting CrewAI LLM start/completion events.
instrumentation/opentelemetry-instrumentation-genai-crewai/tests/conformance/init.py	Adds package marker for conformance scenarios.
instrumentation/opentelemetry-instrumentation-genai-crewai/src/opentelemetry/instrumentation/genai/crewai/event_listener.py	Implements event listener translating CrewAI LLM events into GenAI inference invocations.
instrumentation/opentelemetry-instrumentation-genai-crewai/src/opentelemetry/instrumentation/genai/crewai/init.py	Creates the listener during instrumentation and disables/restores CrewAI native telemetry via env var.
instrumentation/opentelemetry-instrumentation-genai-crewai/README.rst	Updates README to reflect inference spans/metrics and documents native telemetry disabling + completion hooks.
instrumentation/opentelemetry-instrumentation-genai-crewai/examples/custom_hook.py	Adds a minimal custom completion hook example.

[lmolkova]

this is a side-effect we should not do. It's fragile and depends on us modifying env before crewai instrumentation starts. We also can't modify app env. Is there less invasive way?

[lmolkova]

can we be more specific? do we need to read it from both? why getattr(event, "temperature", None) is we know event type?

[lmolkova]

is there more precise type than Any we can use here?

[lmolkova]

there is no point in instrumenting inference calls inside crewai - crewai uses other llm clients to do inference and they are instrumented. if we emit inference spans here, we end up duplicating inference spans from crewai and from corresponding llm client. please remove them from crewai instrumentation.

the spans we should have are invoke_agent, invoke_workflow, tools and planning - https://github.com/open-telemetry/semantic-conventions-genai/blob/main/reference/scenarios/crewai/README.md.