feat: Google GenAI - add FileContent support + refactoring by anakin87 · Pull Request #2860 · deepset-ai/haystack-core-integrations

anakin87 · 2026-02-19T15:07:08Z

Related Issues

fixes Google GenAI - support for FileContent #2828

Proposed Changes:

implement support for FileContent
refactoring:
- splitting implementation and tests across two different files
- use more modern or faster models in tests where possible

How did you test it?

CI

Notes for the reviewer

Unfortunately, the PR is big because I'm moving code around.
In this PR, I've commented the code where the new feature is implemented to make it more visible.

Checklist

I have read the contributors guidelines and the code of conduct
I have updated the related issue with new insights and changes
I added unit tests and updated the docstrings
I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.

anakin87 · 2026-02-19T15:11:17Z

+        elif isinstance(content_part, (ImageContent, FileContent)):
+            cls_name = content_part.__class__.__name__
+
+            if not message.is_from(ChatRole.USER):
+                msg = f"{cls_name} is only supported for user messages"
+                raise ValueError(msg)
+
+            # Validate image MIME type and format
+            if not content_part.mime_type:
+                msg = f"Mime type is required to use {cls_name} with GoogleGenAIChatGenerator"
+                raise ValueError(msg)
+
+            if (
+                isinstance(content_part, ImageContent)
+                and content_part.mime_type not in GOOGLE_GENAI_SUPPORTED_MIME_TYPES
+            ):
+                supported_types = list(GOOGLE_GENAI_SUPPORTED_MIME_TYPES.keys())
+                msg = (
+                    f"Unsupported image MIME type: {content_part.mime_type}. "
+                    f"Google AI supports the following MIME types: {supported_types}"
+                )
+                raise ValueError(msg)
+
+            # Use inline data approach
+            try:
+                base64_data = (
+                    content_part.base64_data if isinstance(content_part, FileContent) else content_part.base64_image
+                )
+                bytes_data = base64.b64decode(base64_data)
+
+                # Create Part using from_bytes method
+                file_part = types.Part.from_bytes(data=bytes_data, mime_type=content_part.mime_type)
+                parts.append(file_part)
+
+            except Exception as e:
+                msg = f"Failed to process {cls_name} data: {e}"
+                raise RuntimeError(msg) from e


this is the new part about FileContent (handled jointly with ImageContent)

anakin87 · 2026-02-19T15:11:46Z

+    def test_convert_message_to_google_genai_file_content(self, test_files_path):
+        file_path = test_files_path / "sample_pdf_3.pdf"
+        file_content = FileContent.from_file_path(file_path)
+        message = ChatMessage.from_user(content_parts=["Describe this document:", file_content])
+        google_content = _convert_message_to_google_genai_format(message)
+        assert google_content.role == "user"
+        assert len(google_content.parts) == 2
+
+        assert google_content.parts[0].text == "Describe this document:"
+        assert hasattr(google_content.parts[1], "inline_data")
+        assert google_content.parts[1].inline_data.mime_type == "application/pdf"
+        assert google_content.parts[1].inline_data.data is not None
+
+    def test_convert_message_to_google_genai_file_content_in_assistant_message(self, test_files_path):
+        file_path = test_files_path / "sample_pdf_3.pdf"
+        file_content = FileContent.from_file_path(file_path)
+        message = ChatMessage.from_assistant("This is a document")
+        message._content.append(file_content)
+
+        with pytest.raises(ValueError, match="FileContent is only supported for user messages"):
+            _convert_message_to_google_genai_format(message)


new unit tests

anakin87 · 2026-02-19T15:12:22Z

+    def test_live_run_with_file_content(self, test_files_path):
+        pdf_path = test_files_path / "sample_pdf_3.pdf"
+
+        file_content = FileContent.from_file_path(file_path=pdf_path)
+
+        chat_messages = [
+            ChatMessage.from_user(
+                content_parts=[file_content, "Is this document a paper about LLMs? Respond with 'yes' or 'no' only."]
+            )
+        ]
+
+        generator = GoogleGenAIChatGenerator()
+        results = generator.run(chat_messages)
+
+        assert len(results["replies"]) == 1
+        message: ChatMessage = results["replies"][0]
+
+        assert message.is_from(ChatRole.ASSISTANT)
+
+        assert message.text
+        assert "no" in message.text.lower()


new integration test

I can imagine this failing because someting changes and the LLM gives variations of a negation, maybe you we could something like this:

indicates_no = any( phrase in message.text.lower() for phrase in ( "no", "nope", "not about", "not a paper about", "it is not", "it's not", "the answer is no", "does not", "doesn't", "negative", ) ) assert indicates_no is True

Done in 15ab505

davidsbatista · 2026-02-23T14:16:47Z

This is going to take a while to review. It would be easier to have this split into two separate PRs, one for the new feature and another for the refactoring.

anakin87 · 2026-02-23T14:42:25Z

This is going to take a while to review. It would be easier to have this split into two separate PRs, one for the new feature and another for the refactoring.

Yes, you are right. Sorry about that.

The only new logic is commented above. The rest is just code moved into separate modules.

If it's okay with you, it would be great to review it as-is. If it turns out to be difficult to review this way, I can split it into two separate PRs.

davidsbatista · 2026-02-23T14:47:03Z

I can review it as is, no worries.

davidsbatista

few small changes/suggestions

davidsbatista · 2026-02-23T15:01:50Z

@@ -830,16 +257,16 @@ def _handle_streaming_response(

            chunk = None


unrelated to the PR - but this chunk = None is never used since the chunk in the loop will always bind to what comes from the response_stream

Good catch!
Removed in 10140a5

davidsbatista · 2026-02-24T10:03:54Z

-

 @component
 class GoogleGenAIChatGenerator:


WE can extend the docstring by saying that user messages can include files (e.g. PDFs) via FileContent. Maybe also add a short example.

Added in 10140a5

davidsbatista · 2026-02-24T10:06:17Z

+                msg = f"{cls_name} is only supported for user messages"
+                raise ValueError(msg)
+
+            # Validate image MIME type and format


nit: consider validation against supported image types instead of only format

I now tried to clarify the comment (10140a5)

davidsbatista · 2026-02-24T10:23:38Z

+    def test_live_run_with_file_content(self, test_files_path):
+        pdf_path = test_files_path / "sample_pdf_3.pdf"
+
+        file_content = FileContent.from_file_path(file_path=pdf_path)
+
+        chat_messages = [
+            ChatMessage.from_user(
+                content_parts=[file_content, "Is this document a paper about LLMs? Respond with 'yes' or 'no' only."]
+            )
+        ]
+
+        generator = GoogleGenAIChatGenerator()
+        results = generator.run(chat_messages)
+
+        assert len(results["replies"]) == 1
+        message: ChatMessage = results["replies"][0]
+
+        assert message.is_from(ChatRole.ASSISTANT)
+
+        assert message.text
+        assert "no" in message.text.lower()


I can imagine this failing because someting changes and the LLM gives variations of a negation, maybe you we could something like this:

indicates_no = any( phrase in message.text.lower() for phrase in ( "no", "nope", "not about", "not a paper about", "it is not", "it's not", "the answer is no", "does not", "doesn't", "negative", ) ) assert indicates_no is True

davidsbatista

LGTM!

feat: Google GenAI - add FileContent support + refactoring

d5c94fc

github-actions Bot added integration:google-genai type:documentation Improvements or additions to documentation labels Feb 19, 2026

anakin87 commented Feb 19, 2026

View reviewed changes

anakin87 added 2 commits February 19, 2026 16:29

revert some changes

095d47c

clean more

813cee2

anakin87 marked this pull request as ready for review February 19, 2026 15:46

anakin87 requested a review from a team as a code owner February 19, 2026 15:46

anakin87 requested review from davidsbatista and removed request for a team February 19, 2026 15:46

davidsbatista reviewed Feb 24, 2026

View reviewed changes

anakin87 added 2 commits February 24, 2026 12:16

Merge branch 'main' into google-filecontent-support

f05c446

implement suggestions

10140a5

anakin87 requested a review from davidsbatista February 24, 2026 11:33

improved test

15ab505

davidsbatista approved these changes Feb 24, 2026

View reviewed changes

anakin87 merged commit 94c9c28 into main Feb 24, 2026
12 checks passed

anakin87 deleted the google-filecontent-support branch February 24, 2026 13:25

anakin87 mentioned this pull request Feb 26, 2026

fix: set include_thoughts=False when thinking_budget is 0 #2853

Merged

4 tasks

		@@ -830,16 +257,16 @@ def _handle_streaming_response(

		chunk = None

Conversation

anakin87 commented Feb 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Related Issues

Proposed Changes:

How did you test it?

Notes for the reviewer

Checklist

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

anakin87 Feb 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

davidsbatista commented Feb 23, 2026

Uh oh!

anakin87 commented Feb 23, 2026

Uh oh!

davidsbatista commented Feb 23, 2026

Uh oh!

davidsbatista left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

anakin87 Feb 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

anakin87 Feb 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

anakin87 Feb 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

davidsbatista left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

anakin87 commented Feb 19, 2026 •

edited

Loading

anakin87 Feb 24, 2026 •

edited

Loading

anakin87 Feb 24, 2026 •

edited

Loading

anakin87 Feb 24, 2026 •

edited

Loading

anakin87 Feb 24, 2026 •

edited

Loading