feat: add multimodal support to OCI Generative AI provider

[fede-kamel]

Summary

• vision.py: vision model lists, to_data_uri, load_image, encode_image, is_vision_model helpers
• _build_generic_content handles image_url, document_url, video_url, audio_url content types → OCI SDK ImageContent, DocumentContent, VideoContent, AudioContent
• _message_has_multimodal_content detects non-text payloads for Cohere rejection
• Cohere models reject multimodal with clear error (text-only)
• supports_multimodal() → True

Depends on #4963, #4962, #4961, #4959. Draft until all merge.
Tracking issue: #4944

Diff breakdown (vs structured output PR)

Change	Lines
vision.py (new file)	+62
completion.py (multimodal content + checks)	+58/-4
__init__.py (vision exports)	+15/-1
test_oci_multimodal.py (10 unit tests)	+195
test_oci_integration_multimodal.py (1 live test)	+106
Total	+440

Test plan

• 10 unit tests: image/document/video/audio content building, unsupported type rejection, Cohere multimodal rejection, multimodal detection, supports_multimodal, vision helpers
• Live integration: sent 2x2 red PNG to google.gemini-2.5-flash via data URI — correctly identified color
• All 33 prior unit tests still pass (no regressions)

[fede-kamel]

OCI GenAI PR series (tracking issue: #4944)

1. feat: add OCI Generative AI provider — basic text completion #4959 — Basic text completion
2. feat: add streaming support to OCI Generative AI provider #4961 — Streaming
3. feat: add tool calling support to OCI Generative AI provider #4962 — Tool calling
4. feat: add structured output support to OCI Generative AI provider #4963 — Structured output
5. feat: add multimodal support to OCI Generative AI provider #4964 — Multimodal ← this PR
6. feat: add OCI Generative AI embeddings provider #4966 — Embeddings
7. feat: add true async support to OCI provider via aiohttp #4982 — True async (aiohttp)

Depends on #4963.

[fede-kamel]

@greysonlalonde @lorenzejay — bumping for review.

Status (after rebase):

• ✅ Rebased onto current main (was 51 commits behind; clean rebase with -X theirs)
• ✅ 0 CI failures, REVIEW_REQUIRED only
• ✅ Built on top of feat: add OCI Generative AI provider — basic text completion #4959 (foundation PR), which now ships the GPT-5-family max_completion_tokens routing fix verified live across 7 model families

This PR adds: Maps crewAI multimodal parts (inline_data Blobs, file_data URLs) to OCI's ImageContent / AudioContent / VideoContent / DocumentContent blocks. Inline bytes are wrapped as data:<mime>;base64,…. Works against Gemini 2.5 vision and Llama 3.2 Vision.

v1 spec compliance: like #4959, this targets OCI's /20231130/Chat v1 API directly via oci.generative_ai_inference. Live HTTP-body capture confirms all spec fields (compartmentId, servingMode, chatRequest{apiFormat,…}) match Oracle's /20231130/Chat reference.

This is part of the 7-PR series (#4959, #4961, #4962, #4963, #4964, #4966, #4982). Happy to fold any subset into a single PR if that helps review velocity.

[coderabbitai]

📝 Walkthrough

Walkthrough

Adds an OCI Generative AI provider with optional SDK installation, native model routing, configurable authentication, provider-specific requests, multimodal inputs, tool calls, structured outputs, synchronous/asynchronous streaming, and unit plus environment-gated integration tests.

Changes

OCI provider integration

Layer / File(s)	Summary
Routing and authentication lib/crewai/pyproject.toml, lib/crewai/src/crewai/llm.py, lib/crewai/src/crewai/utilities/oci.py	Registers the OCI optional dependency, routes OCI model identifiers to OCICompletion, and configures supported OCI authentication modes.
Completion requests and content conversion lib/crewai/src/crewai/llms/providers/oci/*, lib/crewai/tests/llms/oci/test_oci.py, test_oci_cohere_v2.py, test_oci_multimodal.py	Builds generic, Cohere, and Cohere v2 requests, normalizes messages, handles vision content, and applies model-specific request parameters.
Tools and structured responses lib/crewai/src/crewai/llms/providers/oci/completion.py, test_oci_tools.py, test_oci_structured.py	Formats tool calls, executes bounded tool recursion, and parses Pydantic structured responses.
Streaming and public APIs lib/crewai/src/crewai/llms/providers/oci/completion.py, test_oci_streaming.py	Adds synchronous and asynchronous calls, SSE parsing, streamed tool calls, usage metadata, and client locking.
Live integration coverage lib/crewai/tests/llms/oci/conftest.py, test_oci_integration_*.py	Adds environment-gated tests for basic, multimodal, streaming, structured-output, and tool-calling interactions.

Sequence Diagram(s)

sequenceDiagram
  participant CrewAI
  participant OCICompletion
  participant OCI Generative AI
  participant ToolExecutor
  CrewAI->>OCICompletion: call or stream messages
  OCICompletion->>OCI Generative AI: submit provider-specific request
  OCI Generative AI-->>OCICompletion: text, structured output, or tool calls
  OCICompletion->>ToolExecutor: execute available functions
  ToolExecutor-->>OCICompletion: return tool results
  OCICompletion->>OCI Generative AI: submit follow-up request
  OCI Generative AI-->>CrewAI: return final response or streamed chunks

Loading

Suggested reviewers: alex-clawd, lorenzejay, greysonlalonde

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 68.09% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: adding multimodal support to the OCI Generative AI provider.
Description check	✅ Passed	The description matches the changeset and describes the OCI multimodal helpers, content mapping, and tests.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

This PR is stale because it has been open for 45 days with no activity.

[fede-kamel]

Rebased onto current main and un-drafted — mergeable again.

This PR now also adds COHEREV2 api-format support: cohere.command-a-vision rejects the v1 CohereChatRequest, so vision-capable Cohere models route to a new cohere_v2 request family built on CohereChatRequestV2 (message-based history, text + image content, v2 tools, JSON response format, SSE streaming).

Verified live against OCI on-demand (us-chicago-1):

• generic-format vision (meta.llama-4-scout) ✅
• cohere.command-a-vision chat, streaming, and image understanding ✅
• v2 tool calling is wired and unit-tested; the service currently rejects tools for command-a-vision, so it activates when OCI ships a v2 model with tool support.

Unit suite: 53 passed.

[coderabbitai]

Actionable comments posted: 3

🧹 Nitpick comments (4)

lib/crewai/tests/llms/oci/test_oci_streaming.py (1)

10-23: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Move import json to module top level.

import json is inside the function body of _make_fake_stream_event rather than at the module level, which deviates from PEP 8 import placement conventions. As per coding guidelines, Python source files should follow Python best practices.

♻️ Proposed fix

 from __future__ import annotations

+import json
 from unittest.mock import MagicMock

 import pytest


 def _make_fake_stream_event(text: str = "", finish_reason: str | None = None, usage: dict | None = None) -> MagicMock:
     """Build a single SSE event with optional text, finish, and usage."""
     payload: dict = {}
     if text:
         payload["message"] = {"content": [{"text": text}]}
     if finish_reason:
         payload["finishReason"] = finish_reason
     if usage:
         payload["usage"] = usage

-    import json
     event = MagicMock()
     event.data = json.dumps(payload)
     return event

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/tests/llms/oci/test_oci_streaming.py` around lines 10 - 23, Move
the json import from inside _make_fake_stream_event to the module-level imports,
and continue using json.dumps(payload) within the helper.

Source: Coding guidelines

lib/crewai/tests/llms/oci/test_oci_integration_multimodal.py (1)

22-42: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Reuse the shared OCI live-test helpers
_env_models and _skip_unless_live are duplicated in test_oci_integration_tools.py, and conftest.py already has the same live-config flow. This copy has also started to drift (OCI_AUTH_FILE_LOCATION support and skip text differ), so pull the shared env parsing / skip logic into one helper or fixture to keep the OCI integration tests in sync.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/tests/llms/oci/test_oci_integration_multimodal.py` around lines 22
- 42, Replace the duplicated _env_models and _skip_unless_live implementations
with the shared OCI live-test helper or fixture already used by
test_oci_integration_tools.py and conftest.py. Update this test to consume that
shared configuration, preserving support for OCI_AUTH_FILE_LOCATION and
consistent skip messages, and remove the local helper definitions.

lib/crewai/src/crewai/llms/providers/oci/vision.py (1)

46-57: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Add docstrings for public API functions.

load_image, encode_image, and is_vision_model are exported in __all__ but lack docstrings, while to_data_uri has one. As per coding guidelines, public APIs should be documented.

♻️ Proposed fix

 def load_image(file_path: str | Path) -> dict[str, dict[str, str] | str]:
+    """Load an image from a file path and return an image_url content dict."""
     return {"type": "image_url", "image_url": {"url": to_data_uri(file_path)}}


 def encode_image(
     image_bytes: bytes, mime_type: str = "image/png"
 ) -> dict[str, dict[str, str] | str]:
+    """Encode raw image bytes into an image_url content dict with a data URI."""
     return {"type": "image_url", "image_url": {"url": to_data_uri(image_bytes, mime_type)}}


 def is_vision_model(model_id: str) -> bool:
+    """Return True if the model supports vision/multimodal inputs."""
     return model_id in VISION_MODELS

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/src/crewai/llms/providers/oci/vision.py` around lines 46 - 57, Add
concise docstrings to the public functions load_image, encode_image, and
is_vision_model, documenting their parameters, return values, and behavior
consistently with the existing to_data_uri documentation.

Source: Coding guidelines

lib/crewai/tests/llms/oci/test_oci_multimodal.py (1)

10-104: 📐 Maintainability & Code Quality | 🔵 Trivial | 💤 Low value

Strengthen content-building tests with argument verification.

The image, document, video, and audio content tests verify that the correct OCI constructors are called (assert_called_once()) but don't verify the arguments passed (e.g., URL values). Adding assert_called_once_with(...) would catch bugs where the wrong URL is passed to the constructor.

♻️ Example improvement for test_oci_builds_image_content

     assert len(result) == 2
     patch_oci_module.generative_ai_inference.models.ImageContent.assert_called_once()
-    patch_oci_module.generative_ai_inference.models.ImageUrl.assert_called_once()
+    patch_oci_module.generative_ai_inference.models.ImageUrl.assert_called_once_with(
+        url="data:image/png;base64,abc123"
+    )

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/tests/llms/oci/test_oci_multimodal.py` around lines 10 - 104,
Strengthen the multimodal content tests by replacing constructor-only call-count
assertions with argument verification. In test_oci_builds_image_content,
test_oci_builds_document_content, test_oci_builds_video_content, and
test_oci_builds_audio_content, assert ImageUrl, DocumentUrl, VideoUrl, and
AudioUrl are each called once with the corresponding input URL and assert the
content constructors receive the expected URL object.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@lib/crewai/pyproject.toml`:
- Around line 107-109: Raise the minimum OCI dependency in the `oci` optional
extra from `2.168.0` to `2.168.2` so installations receive the security-patched
urllib3 release.

In `@lib/crewai/src/crewai/llms/providers/oci/completion.py`:
- Around line 1371-1375: Update the streaming path around `_chat()` and
`response.data.events()` so `_client_lock` remains held for the entire
event-consumption lifetime, not just while creating the response; mirror the
locking pattern used by `_stream_chat_events()` and ensure the lock is released
via an idiomatic context manager even if iteration raises.

In `@lib/crewai/src/crewai/utilities/oci.py`:
- Around line 46-47: Update the OCI security-token configuration handling in the
relevant function to validate that `key_file` and `security_token_file` exist
before accessing them. Raise a descriptive configuration error identifying the
missing SECURITY_TOKEN entries instead of allowing a raw `KeyError` to escape.

---

Nitpick comments:
In `@lib/crewai/src/crewai/llms/providers/oci/vision.py`:
- Around line 46-57: Add concise docstrings to the public functions load_image,
encode_image, and is_vision_model, documenting their parameters, return values,
and behavior consistently with the existing to_data_uri documentation.

In `@lib/crewai/tests/llms/oci/test_oci_integration_multimodal.py`:
- Around line 22-42: Replace the duplicated _env_models and _skip_unless_live
implementations with the shared OCI live-test helper or fixture already used by
test_oci_integration_tools.py and conftest.py. Update this test to consume that
shared configuration, preserving support for OCI_AUTH_FILE_LOCATION and
consistent skip messages, and remove the local helper definitions.

In `@lib/crewai/tests/llms/oci/test_oci_multimodal.py`:
- Around line 10-104: Strengthen the multimodal content tests by replacing
constructor-only call-count assertions with argument verification. In
test_oci_builds_image_content, test_oci_builds_document_content,
test_oci_builds_video_content, and test_oci_builds_audio_content, assert
ImageUrl, DocumentUrl, VideoUrl, and AudioUrl are each called once with the
corresponding input URL and assert the content constructors receive the expected
URL object.

In `@lib/crewai/tests/llms/oci/test_oci_streaming.py`:
- Around line 10-23: Move the json import from inside _make_fake_stream_event to
the module-level imports, and continue using json.dumps(payload) within the
helper.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: b0ef515d-fc37-4df1-b0f7-82d596e458b5

📥 Commits

Reviewing files that changed from the base of the PR and between 85c467d and 8abc166.

📒 Files selected for processing (19)

• lib/crewai/pyproject.toml
• lib/crewai/src/crewai/llm.py
• lib/crewai/src/crewai/llms/providers/oci/__init__.py
• lib/crewai/src/crewai/llms/providers/oci/completion.py
• lib/crewai/src/crewai/llms/providers/oci/vision.py
• lib/crewai/src/crewai/utilities/oci.py
• lib/crewai/tests/llms/oci/__init__.py
• lib/crewai/tests/llms/oci/conftest.py
• lib/crewai/tests/llms/oci/test_oci.py
• lib/crewai/tests/llms/oci/test_oci_cohere_v2.py
• lib/crewai/tests/llms/oci/test_oci_integration_basic.py
• lib/crewai/tests/llms/oci/test_oci_integration_multimodal.py
• lib/crewai/tests/llms/oci/test_oci_integration_streaming.py
• lib/crewai/tests/llms/oci/test_oci_integration_structured.py
• lib/crewai/tests/llms/oci/test_oci_integration_tools.py
• lib/crewai/tests/llms/oci/test_oci_multimodal.py
• lib/crewai/tests/llms/oci/test_oci_streaming.py
• lib/crewai/tests/llms/oci/test_oci_structured.py
• lib/crewai/tests/llms/oci/test_oci_tools.py

[coderabbitai]

🔒 Security & Privacy | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🌐 Web query:

What is the latest version of the oci Python SDK on PyPI, and are there any known security advisories for versions >=2.168.0?

💡 Result:

The latest version of the Oracle Cloud Infrastructure (OCI) Python SDK available on PyPI as of July 10, 2026, is 2.181.0 [1][2]. Regarding security advisories for versions >= 2.168.0, the SDK itself is typically not the source of direct vulnerabilities but occasionally requires dependency updates (such as for the 'cryptography' or 'urllib3' libraries) to address vulnerabilities found in those upstream packages [3][4][5][6]. For instance, version 2.168.2 included a specific security update to address CVE-2025-66418 by adjusting the upper bound of the 'urllib3' dependency for Python >= 3.10 [3]. Additionally, versions in the 2.168.x range and subsequent releases have addressed previous concerns regarding dependency version constraints (specifically for the 'cryptography' library) that had previously prevented users from applying security patches to those underlying libraries [4]. Oracle manages security for the SDK through regular release cadences [7]. If you are using version 2.168.0 or later, it is recommended to keep the SDK updated to the latest version (2.181.0) to ensure you have the most recent dependency constraints and patches [8][4]. You can track specific vulnerability reports and security updates through the project's official GitHub repository [9][7].

Citations:

• 1: https://pypi.org/project/oci/
• 2: https://pypistats.org/packages/oci
• 3: https://github.com/oracle/oci-python-sdk/releases/tag/v2.168.2
• 4: Cannot take security patch for Cryptography library due to upper bound on cryptography version oracle/oci-python-sdk#829
• 5: Cryptography security vulnerabilities CVE-2023-50782, CVE-2023-5678, CVE-2023-6129, CVE-2023-6237 oracle/oci-python-sdk#618
• 6: Upgrade cryptography Version to >=43.0.1 Due to High Severity Security Vulnerability oracle/oci-python-sdk#700
• 7: https://github.com/oracle/oci-python-sdk/security
• 8: https://docs.oracle.com/en-us/iaas/tools/python/latest/installation.html
• 9: https://github.com/oracle/oci-python-sdk/releases

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

# Inspect the pyproject section and OCI usage to judge whether the minimum pin is justified.
printf '\n## pyproject excerpt\n'
sed -n '95,120p' lib/crewai/pyproject.toml

printf '\n## oci imports/usages\n'
rg -n '\bimport oci\b|\bfrom oci\b|oci\.' lib/crewai -g '!**/__pycache__/**'

printf '\n## files mentioning optional dependency or auth helpers\n'
rg -n 'optional-dependencies|extras|oci' lib/crewai -g 'pyproject.toml' -g '*.py'

Repository: crewAIInc/crewAI

Length of output: 50372

---

Raise the OCI minimum to 2.168.2 oci>=2.168.0 misses the urllib3 security-update release; bump the optional extra so users get the patched dependency floor.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/pyproject.toml` around lines 107 - 109, Raise the minimum OCI
dependency in the `oci` optional extra from `2.168.0` to `2.168.2` so
installations receive the security-patched urllib3 release.

[coderabbitai]

🩺 Stability & Availability | 🟠 Major | ⚡ Quick win

Keep the client lock for the full iter_stream lifetime.

_chat() releases _client_lock before response.data.events() is consumed. A concurrent call can therefore enter client.chat while this stream is active, unlike _stream_chat_events() which correctly holds the lock throughout event consumption.

Proposed fix

-        response = self._chat(chat_details)
         usage_data: dict[str, int] = {}
         response_metadata: dict[str, Any] = {}

-        for event in response.data.events():
+        for event in self._stream_chat_events(chat_details):

As per coding guidelines, “Follow Python best practices and idiomatic patterns in Python source files.”

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	response = self._chat(chat_details)
	usage_data: dict[str, int] = {}
	response_metadata: dict[str, Any] = {}
	for event in response.data.events():
	usage_data: dict[str, int] = {}
	response_metadata: dict[str, Any] = {}
	for event in self._stream_chat_events(chat_details):

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/src/crewai/llms/providers/oci/completion.py` around lines 1371 -
1375, Update the streaming path around `_chat()` and `response.data.events()` so
`_client_lock` remains held for the entire event-consumption lifetime, not just
while creating the response; mirror the locking pattern used by
`_stream_chat_events()` and ensure the lock is released via an idiomatic context
manager even if iteration raises.

Source: Coding guidelines

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Guard against KeyError on SECURITY_TOKEN config access.

Direct dict access config["key_file"] and config["security_token_file"] will raise a cryptic KeyError if the OCI config profile is missing these entries. A descriptive error improves UX for misconfigured security token auth.

🛡️ Proposed fix

         key_file = config.get("key_file")
         if not key_file:
             raise ValueError(
                 "OCI SECURITY_TOKEN auth requires 'key_file' in the config profile"
             )
         security_token_file = config.get("security_token_file")
         if not security_token_file:
             raise ValueError(
                 "OCI SECURITY_TOKEN auth requires 'security_token_file' in the config profile"
             )

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	key_file = config["key_file"]
	security_token_file = config["security_token_file"]
	key_file = config.get("key_file")
	if not key_file:
	raise ValueError(
	"OCI SECURITY_TOKEN auth requires 'key_file' in the config profile"
	)
	security_token_file = config.get("security_token_file")
	if not security_token_file:
	raise ValueError(
	"OCI SECURITY_TOKEN auth requires 'security_token_file' in the config profile"
	)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/src/crewai/utilities/oci.py` around lines 46 - 47, Update the OCI
security-token configuration handling in the relevant function to validate that
`key_file` and `security_token_file` exist before accessing them. Raise a
descriptive configuration error identifying the missing SECURITY_TOKEN entries
instead of allowing a raw `KeyError` to escape.

[coderabbitai]

Actionable comments posted: 1

♻️ Duplicate comments (1)

lib/crewai/src/crewai/llms/providers/oci/completion.py (1)

1371-1375: 🩺 Stability & Availability | 🟠 Major

Keep the client lock for the full iter_stream lifetime.

_chat() releases _client_lock before response.data.events() is consumed. A concurrent call can enter client.chat while this stream is active, unlike _stream_chat_events() which correctly holds the lock throughout event consumption.

Proposed fix

-        response = self._chat(chat_details)
         usage_data: dict[str, int] = {}
         response_metadata: dict[str, Any] = {}

-        for event in response.data.events():
+        for event in self._stream_chat_events(chat_details):

As per coding guidelines, "Follow Python best practices and idiomatic patterns in Python source files."

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/src/crewai/llms/providers/oci/completion.py` around lines 1371 -
1375, Update the streaming path around `_chat()` and `response.data.events()` so
`_client_lock` remains acquired for the entire event-iteration lifetime,
matching `_stream_chat_events()`; move or restructure lock acquisition to
encompass the client call and all consumption of `response.data.events()`,
releasing it only after iteration completes, including exception paths.

Source: Coding guidelines

🧹 Nitpick comments (1)

lib/crewai/tests/llms/oci/test_oci_tools.py (1)

148-205: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Add tests for Cohere v2 tool formatting and extraction.

_format_tools has a cohere_v2 branch (producing CohereToolV2) and _extract_tool_calls has a cohere_v2 branch (using _cohere_v2_function_field), but neither path is tested. These are important new code paths for the COHEREV2 support described in the PR objectives. As per coding guidelines, "Write unit tests for new functionality that focus on behavior rather than implementation details."

Suggested test additions

def test_oci_completion_formats_cohere_v2_tools(
    patch_oci_module, oci_unit_values
):
    """_format_tools should produce CohereToolV2 for Cohere v2 models."""
    from crewai.llms.providers.oci.completion import OCICompletion

    patch_oci_module.generative_ai_inference.GenerativeAiInferenceClient.return_value = MagicMock()
    llm = OCICompletion(
        model="cohere.command-a-vision",
        compartment_id=oci_unit_values["compartment_id"],
    )
    formatted = llm._format_tools(SAMPLE_TOOLS)

    assert len(formatted) == 1
    models = patch_oci_module.generative_ai_inference.models
    models.CohereToolV2.assert_called_once()


def _make_cohere_v2_tool_call_response(tool_name: str = "get_weather") -> MagicMock:
    """Build a fake OCI Cohere v2 response with a tool call."""
    tc = MagicMock()
    tc.id = "tc_v2_001"
    function = MagicMock()
    function.name = tool_name
    function.arguments = '{"city": "NYC"}'
    tc.function = function

    message = MagicMock()
    message.tool_calls = [tc]

    chat_response = MagicMock()
    chat_response.message = message
    chat_response.text = ""
    chat_response.finish_reason = "COMPLETE"
    chat_response.usage = MagicMock(prompt_tokens=8, completion_tokens=4, total_tokens=12)

    response = MagicMock()
    response.data.chat_response = chat_response
    return response


def test_oci_completion_cohere_v2_extracts_tool_calls(
    patch_oci_module, oci_unit_values
):
    """Cohere v2 tool calls should be normalized with function field extraction."""
    from crewai.llms.providers.oci.completion import OCICompletion

    fake_client = MagicMock()
    fake_client.chat.return_value = _make_cohere_v2_tool_call_response()
    patch_oci_module.generative_ai_inference.GenerativeAiInferenceClient.return_value = fake_client

    llm = OCICompletion(
        model="cohere.command-a-vision",
        compartment_id=oci_unit_values["compartment_id"],
    )
    result = llm.call(
        messages=[{"role": "user", "content": "Weather?"}],
        tools=SAMPLE_TOOLS,
    )

    assert isinstance(result, list)
    assert result[0]["function"]["name"] == "get_weather"
    assert result[0]["id"] == "tc_v2_001"

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/tests/llms/oci/test_oci_tools.py` around lines 148 - 205, Add unit
coverage in the OCI completion tests for both Cohere v2 tool paths: create a
test for _format_tools using a Cohere v2 model and assert it returns a tool via
models.CohereToolV2, and add a helper response fixture plus a test for
_extract_tool_calls through OCICompletion.call that verifies the normalized
function name and preserves the tool-call ID.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@lib/crewai/src/crewai/llms/providers/oci/completion.py`:
- Around line 1222-1227: In _call_impl, pass normalized_messages rather than
messages to _finalize_text_response, matching _stream_call_impl and the
preceding _parse_structured_response call so downstream hooks receive
list[LLMMessage] consistently.

---

Duplicate comments:
In `@lib/crewai/src/crewai/llms/providers/oci/completion.py`:
- Around line 1371-1375: Update the streaming path around `_chat()` and
`response.data.events()` so `_client_lock` remains acquired for the entire
event-iteration lifetime, matching `_stream_chat_events()`; move or restructure
lock acquisition to encompass the client call and all consumption of
`response.data.events()`, releasing it only after iteration completes, including
exception paths.

---

Nitpick comments:
In `@lib/crewai/tests/llms/oci/test_oci_tools.py`:
- Around line 148-205: Add unit coverage in the OCI completion tests for both
Cohere v2 tool paths: create a test for _format_tools using a Cohere v2 model
and assert it returns a tool via models.CohereToolV2, and add a helper response
fixture plus a test for _extract_tool_calls through OCICompletion.call that
verifies the normalized function name and preserves the tool-call ID.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: a33013ea-cb76-49eb-a68f-8e3272986042

📥 Commits

Reviewing files that changed from the base of the PR and between 8abc166 and f68fc6e.

📒 Files selected for processing (14)

• lib/crewai/src/crewai/llms/providers/oci/__init__.py
• lib/crewai/src/crewai/llms/providers/oci/completion.py
• lib/crewai/src/crewai/llms/providers/oci/vision.py
• lib/crewai/tests/llms/oci/conftest.py
• lib/crewai/tests/llms/oci/test_oci_cohere_v2.py
• lib/crewai/tests/llms/oci/test_oci_integration_basic.py
• lib/crewai/tests/llms/oci/test_oci_integration_multimodal.py
• lib/crewai/tests/llms/oci/test_oci_integration_streaming.py
• lib/crewai/tests/llms/oci/test_oci_integration_structured.py
• lib/crewai/tests/llms/oci/test_oci_integration_tools.py
• lib/crewai/tests/llms/oci/test_oci_multimodal.py
• lib/crewai/tests/llms/oci/test_oci_streaming.py
• lib/crewai/tests/llms/oci/test_oci_structured.py
• lib/crewai/tests/llms/oci/test_oci_tools.py

✅ Files skipped from review due to trivial changes (1)

• lib/crewai/src/crewai/llms/providers/oci/init.py

🚧 Files skipped from review as they are similar to previous changes (9)

• lib/crewai/tests/llms/oci/test_oci_integration_basic.py
• lib/crewai/tests/llms/oci/test_oci_integration_tools.py
• lib/crewai/src/crewai/llms/providers/oci/vision.py
• lib/crewai/tests/llms/oci/test_oci_integration_structured.py
• lib/crewai/tests/llms/oci/test_oci_integration_streaming.py
• lib/crewai/tests/llms/oci/test_oci_integration_multimodal.py
• lib/crewai/tests/llms/oci/test_oci_cohere_v2.py
• lib/crewai/tests/llms/oci/conftest.py
• lib/crewai/tests/llms/oci/test_oci_multimodal.py

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Pass normalized_messages instead of messages to _finalize_text_response.

_call_impl passes the original messages (which may be a string) to _finalize_text_response, while _stream_call_impl (line 1349) correctly passes normalized_messages. The _parse_structured_response call at line 1217 also correctly uses normalized_messages, confirming this is an oversight. If _call_impl is ever called directly with a string, _finalize_text_response and its downstream hooks/events would receive a string instead of the expected list[LLMMessage].

Proposed fix

         return self._finalize_text_response(
             content=content,
-            messages=messages,
+            messages=normalized_messages,
             from_task=from_task,
             from_agent=from_agent,
         )

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	return self._finalize_text_response(
	content=content,
	messages=messages,
	from_task=from_task,
	from_agent=from_agent,
	)
	return self._finalize_text_response(
	content=content,
	messages=normalized_messages,
	from_task=from_task,
	from_agent=from_agent,
	)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@lib/crewai/src/crewai/llms/providers/oci/completion.py` around lines 1222 -
1227, In _call_impl, pass normalized_messages rather than messages to
_finalize_text_response, matching _stream_call_impl and the preceding
_parse_structured_response call so downstream hooks receive list[LLMMessage]
consistently.