feat: Integrations core classes by akihikokuroda · Pull Request #24 · generative-computing/mellea-contribs

akihikokuroda · 2026-02-17T18:48:07Z

Fix: generative-computing/mellea#447

mergify · 2026-02-17T18:48:42Z

Merge Protections

Your pull request matches the following merge protections and will not be merged until they are valid.

🟢 Enforce conventional commit

Wonderful, this rule succeeded.

Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/

title ~= ^(fix|feat|docs|style|refactor|perf|test|build|ci|chore|revert|release)(?:\(.+\))?:

psschwei

Couple of comments here:

We should confirm with @nrfulton where we should put the integrations. My understanding is that mellea-contribs is going to be a single package, but let's confirm. (if that is the case, we'll need to combine the pyproject.tomls)
10 of the tests failed when I tried running them. Here's how I ran them:

cd integrations/mellea-integration-core
uv sync
uv add pytest
uv run pytest

and the failing tests:

FAILED tests/test_base.py::test_prepare_generation_no_messages - Failed: DID NOT RAISE <class 'ValueError'>
FAILED tests/test_base.py::test_agenerate_with_mellea_chat - Failed: async def functions are not natively supported.
FAILED tests/test_base.py::test_agenerate_with_mellea_instruct - Failed: async def functions are not natively supported.
FAILED tests/test_base.py::test_full_agenerate_flow - Failed: async def functions are not natively supported.
FAILED tests/test_message_converter.py::test_extract_content_from_response_fallback - AttributeError: Mock object has no attribute '__str__'
FAILED tests/test_tool_converter.py::test_extract_tool_schema_basic - AssertionError: assert <Mock name='test_tool.name' id='4418542224'> == 'test_tool'
FAILED tests/test_tool_converter.py::test_get_tool_callable_not_found - Failed: DID NOT RAISE <class 'ValueError'>
FAILED tests/test_tool_converter.py::test_extract_tool_calls_from_response_tool_calls_attr - AssertionError: assert <Mock name='test_tool.name' id='4418765072'> == 'test_tool'
FAILED tests/test_tool_converter.py::test_extract_tool_calls_from_response_private_attr - AssertionError: assert <Mock name='test_tool.name' id='4418769968'> == 'test_tool'
FAILED tests/test_tool_converter.py::test_parse_tool_calls_default_implementation - AssertionError: assert <Mock name='test_tool.name' id='4418772944'> == 'test_tool'

related point, we should add these tests to the CI

akihikokuroda · 2026-03-06T20:23:05Z

I might forget check these tests. Thanks for checking.

akihikokuroda · 2026-03-06T20:25:54Z

Tests can be added once we can confirm home of these integrations.

psschwei · 2026-03-06T21:05:30Z

+- ✅ **Tool Conversion**: Utilities for handling tool calling across frameworks
+- ✅ **Requirements Support**: Built-in support for Mellea's validation features
+- ✅ **Async Support**: Both sync and async generation patterns
+- ✅ **Type Safety**: Protocol-based design with type hints


While we use type hints, a lot of them are Any which isn't exactly type safe

Adding some clarification here.

psschwei · 2026-03-06T21:08:45Z

+        valid_roles = ["system", "user", "assistant", "tool"]
+        if role not in valid_roles:
+            role = "user"  # Default fallback


should we instead log an error, since if the role isn't valid that may mean we have a bug in the conversion logic ?

psschwei · 2026-03-06T21:13:08Z

+            ValueError: If validation failed and no samples were generated
+        """
+        # Check if this is a SamplingResult object
+        if hasattr(response, "success") and hasattr(response, "result"):


Would it be better to explicitly check, something like

Suggested change

if hasattr(response, "success") and hasattr(response, "result"):

if SamplingResult is not None and isinstance(response, SamplingResult):

?

psschwei · 2026-03-06T21:15:41Z

+        else:
+            # If response has no content or value attribute, return empty string
+            # rather than string representation of the object
+            return ""


would it be worth logging this (assuming an empty response could be a sign of something having gone wrong)?

psschwei · 2026-03-06T21:22:25Z

+
+        # Pattern to match ToolCall objects in the string
+        # Matches: ToolCall(function=Function(name='tool_name', arguments={...}))
+        pattern = r"ToolCall\(function=Function\(name='([^']+)',\s*arguments=(\{[^}]+\})\)\)"


is there any chance that arguments could be a nested dict? e.g. {'a': {'b': 1}} if so, I think the regex may fail on that case

Updated to accept the nested case. Comments are added for clarification.

psschwei · 2026-03-06T21:23:19Z

+## Existing Integrations
+
+This core package is used by:
+
+- **[mellea-langchain](../langchain/)**: LangChain integration
+- **[mellea-crewai](../crewai/)**: CrewAI integration
+- **[mellea-dspy](../dspy/)**: DSPy integration


we may want to hold off adding these until we land those integrations

psschwei · 2026-03-10T15:08:59Z

+        # Note: This pattern uses a greedy match to handle nested dictionaries
+        # It will match from the first { to the last } in the arguments section
+        pattern = r"ToolCall\(function=Function\(name='([^']+)',\s*arguments=(\{.+?\})\)\)"


technically, .+? is lazy not greedy. for greedy, you'd need just.+
it might be worth doing something like regex-extracting just the tool name, finding the arguments prefix, and then using a brace-counting scan (a small helper to match closing and opening braces) to grab the full dict string ? wdyt?

Thanks for suggestion. I implemented it.

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

psschwei

I think I'm generally ok with this, pending resolution of the location question.
@AngeloDanducci can you also give a review?

AngeloDanducci · 2026-03-13T07:09:22Z

+            try:
+                schema["function"]["parameters"] = tool.args_schema.schema()
+            except Exception:
+                # If schema extraction fails, use empty dict


May be worth logging the exception, doesn't seem like a big deal though.

AngeloDanducci · 2026-03-13T07:18:46Z

Had a couple of failures in here

FAILED tests/test_base.py::test_handle_sampling_results_success - AssertionError: assert <Mock id='4426200416'> == <Mock name='mock.result' id='4426200128'> FAILED tests/test_base.py::test_handle_sampling_results_failure_with_samples - AssertionError: assert <Mock id='4426102064'> == <Mock id='4426097168'> FAILED tests/test_base.py::test_handle_sampling_results_failure_no_samples - Failed: DID NOT RAISE <class 'ValueError'>

AngeloDanducci · 2026-03-13T07:21:54Z

Couple of test failures but otherwise LGTM.

Re location - I think it makes sense that each "sub project" could have its own toml but it's not a huge deal to me either way.

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

akihikokuroda · 2026-03-13T12:23:57Z

Thanks for review. I made fixes and suggested changes.

AngeloDanducci

Per offline discussion, we should likely move to uv workspaces for the approach of multi interconnected packages in the same repo.

If you want to set that up in this PR great if not feel free to merge and I'll open an issue to follow up after we get the integrations core PRs in.

akihikokuroda marked this pull request as draft February 17, 2026 18:48

akihikokuroda marked this pull request as ready for review February 27, 2026 18:18

psschwei reviewed Mar 6, 2026

View reviewed changes

akihikokuroda requested a review from psschwei March 6, 2026 22:16

psschwei reviewed Mar 10, 2026

View reviewed changes

akihikokuroda added 5 commits March 11, 2026 10:28

feat: integration base classes

1e34048

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

after langchain refactor

ecc1713

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

fix test errors

5cf1176

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

review comments and fix format issue

e979efc

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

review comment

662b1dd

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

akihikokuroda force-pushed the refactor branch from 5604d1c to 662b1dd Compare March 11, 2026 14:28

psschwei reviewed Mar 11, 2026

View reviewed changes

AngeloDanducci reviewed Mar 13, 2026

View reviewed changes

This was referenced Mar 13, 2026

feat: refactor crewai with integration core #26

Merged

feat: crewai guardrail integration #30

Merged

review comments

181fb22

Signed-off-by: Akihiko Kuroda <akihikokuroda2020@gmail.com>

AngeloDanducci approved these changes Mar 13, 2026

View reviewed changes

akihikokuroda merged commit 50d2dd8 into generative-computing:main Mar 13, 2026
4 checks passed

akihikokuroda deleted the refactor branch March 13, 2026 17:54

	if hasattr(response, "success") and hasattr(response, "result"):
	if SamplingResult is not None and isinstance(response, SamplingResult):

Conversation

akihikokuroda commented Feb 17, 2026

Uh oh!

mergify Bot commented Feb 17, 2026

Merge Protections

🟢 Enforce conventional commit

Uh oh!

psschwei left a comment

Choose a reason for hiding this comment

Uh oh!

akihikokuroda commented Mar 6, 2026

Uh oh!

akihikokuroda commented Mar 6, 2026

Uh oh!

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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

akihikokuroda Mar 6, 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!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

psschwei left a comment

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!

AngeloDanducci commented Mar 13, 2026

Uh oh!

akihikokuroda commented Mar 13, 2026

Uh oh!

AngeloDanducci left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

akihikokuroda Mar 6, 2026 •

edited

Loading