Enh/ap 25561 update bundled env to py 3 13 #42

marc-lehner · 2026-02-09T11:00:10Z

No description provided.

Copilot

Pull request overview

Updates the bundled environment and LangChain integration to support Python 3.13, including migration to newer/alternate LangChain packages and fixes for evaluation tooling compatibility.

Changes:

Migrates several LangChain imports/usages to langchain_classic, langchain_core, and langgraph APIs.
Updates Google model integrations to use langchain_google_genai and adjusts API base URL handling.
Modernizes the bundled environment (Pixi) for Python 3.13 and adds a SciPy monkey-patch to unblock Giskard imports.

Reviewed changes

Copilot reviewed 11 out of 12 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
src/tools/vectorstore.py	Switches retrieval chain imports to `langchain_classic`.
src/models/google/_port_types.py	Updates Google chat/embedding models to `langchain_google_genai` and adjusts base URL scheme.
src/models/_adapter.py	Updates LLM invocation to use `invoke()`.
src/indexes/chroma.py	Changes Chroma persistence handling and clarifies hack comment.
src/eval/giskard/giskard_patch.py	Adds SciPy monkey-patch to fix Giskard import compatibility.
src/eval/giskard/init.py	Ensures monkey-patch is imported/applied early for Giskard.
src/agents/openai.py	Switches agent factory import to `langchain_classic`.
src/agents/base_deprecated.py	Updates langgraph react agent wiring and recursion limit handling logic.
src/agents/base.py	Migrates execution path to langgraph react agent and tool node usage.
src/agents/_tool.py	Switches `StructuredTool` import to `langchain_core.tools`.
pixi.toml	Reworks dependencies and bumps runtime to Python 3.13.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-09T11:02:45Z

src/agents/base_deprecated.py

+        from ._agent import RECURSION_CONTINUE_PROMPT
+        if self._recursion_limit_handling == RecursionLimitModeForView.CONFIRM.name:
            message = {
                "type": "ai",
                "content": ITERATION_CONTINUE_PROMPT,


RecursionLimitModeForView and ITERATION_CONTINUE_PROMPT are referenced but (based on this diff) are not imported/defined here, while RECURSION_CONTINUE_PROMPT is imported but not used. This will raise NameError at runtime. Make the identifiers consistent (use the newly imported RECURSION_CONTINUE_PROMPT for the message content and append-to-memory call), and either import the correct RecursionLimitModeForView enum or keep using the existing IterationLimitModeForView enum already imported in this module.

Copilot · 2026-02-09T11:02:46Z

src/agents/base.py

+        if self.user_message:
+            messages.append({"role": "user", "content": self.user_message})


This appends a raw dict into messages, but earlier the code builds messages from to_langchain_message(...) (LangChain BaseMessage instances). Mixing dicts and BaseMessage objects is likely to break downstream LangGraph/LangChain processing (e.g., attribute access on message objects). Append a proper LangChain message object instead (e.g., langchain_core.messages.HumanMessage(content=...)) to keep types consistent.

src/agents/base.py

Copilot · 2026-02-09T11:02:47Z

src/eval/giskard/giskard_patch.py

+            pass
+
+
+apply_patch()


Calling apply_patch() at import time creates unavoidable side effects, and the patch is already triggered by importing this module in src/eval/giskard/__init__.py. Prefer making apply_patch() idempotent and calling it explicitly from __init__.py (or gating the call behind a deliberate entrypoint) so imports remain predictable and testing/mocking is easier.

Suggested change

apply_patch()

if __name__ == "__main__":

apply_patch()

Copilot · 2026-02-09T11:02:47Z

src/indexes/chroma.py

@@ -202,43 +202,39 @@ def save_vectorstore(self, vectorstore_folder: str, vectorstore):
        from langchain_chroma import Chroma



The previous guard that avoided the hack when the persist directory already matched was removed. If the subsequent hack logic is expensive (e.g., rebuilding a client or rewriting persistence), this can become an always-on cost. If the new langchain_chroma.Chroma no longer exposes _persist_directory, consider reintroducing a lightweight guard using a supported/public way to read the underlying persist directory (e.g., via the Chroma client/settings) and only run the hack when it’s actually needed.

Suggested change

# If the vectorstore already persists to the desired directory, skip the expensive rebuild.

existing_persist_dir = getattr(vectorstore, "persist_directory", None)

if existing_persist_dir is not None and existing_persist_dir == vectorstore_folder:

return

Copilot · 2026-02-09T11:02:47Z

src/models/google/_port_types.py

    @property
    def base_api_url(self) -> str:
-        return f"{self.location}-aiplatform.googleapis.com"
+        return f"https://{self.location}-aiplatform.googleapis.com"


base_api_url now always includes the scheme (https://...), but custom_base_api_url may still be provided without a scheme (depending on how it’s configured/serialized elsewhere). In that case, base_url could become an invalid URL. Consider normalizing custom_base_api_url to ensure it includes a scheme (or documenting/enforcing that it must be a full URL) so base_url is consistently well-formed.

Copilot · 2026-02-09T11:02:48Z

src/models/google/_port_types.py

            max_tokens=max_tokens,
            temperature=temperature,
            max_retries=2,  # default is 6, instead we just try twice before failing
            base_url=self.custom_base_api_url or self.base_api_url,


base_api_url now always includes the scheme (https://...), but custom_base_api_url may still be provided without a scheme (depending on how it’s configured/serialized elsewhere). In that case, base_url could become an invalid URL. Consider normalizing custom_base_api_url to ensure it includes a scheme (or documenting/enforcing that it must be a full URL) so base_url is consistently well-formed.

Copilot

Pull request overview

Copilot reviewed 11 out of 12 changed files in this pull request and generated 3 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-11T10:48:28Z

src/models/google/_port_types.py

            max_retries=2,  # default is 6, instead we just try twice before failing
            base_url=self.custom_base_api_url or self.base_api_url,
            credentials=google_credentials,
+            task_type="RETRIEVAL_QUERY", # for backwards compatibility with old VertexAIEmbeddings


Missing space after '#' in inline comment. Should be '# for backwards compatibility with old VertexAIEmbeddings'

Copilot · 2026-02-11T10:48:29Z

pixi.toml

+
+# --- Vector Stores & Indexes ---
+faiss-cpu = "*" # Direct: used in indexes/faiss.py
+langchain-chroma = "==1.1.0" # Direct: used in indexes/chroma.py installed via pypi because conda packages don't support python 3.13


The comment is overly long and should be split. Consider moving 'installed via pypi because conda packages don't support python 3.13' to a separate comment line above for better readability.

Suggested change

langchain-chroma = "==1.1.0" # Direct: used in indexes/chroma.py installed via pypi because conda packages don't support python 3.13

# Installed via PyPI because conda packages don't support Python 3.13

langchain-chroma = "==1.1.0" # Direct: used in indexes/chroma.py

Copilot · 2026-02-11T10:48:29Z

src/agents/base_deprecated.py

+        tool_node = ToolNode(tools, handle_tool_errors=True)
        agent = create_react_agent(
-            chat_model, tools=tools, prompt=self.developer_message, checkpointer=memory
+            chat_model, tools=tool_node, prompt=self.developer_message, checkpointer=memory


The API change passes a ToolNode instance to the 'tools' parameter instead of a list of tools. This breaks the parameter naming convention and could confuse API consumers. Consider renaming the parameter or documenting this breaking change clearly.

Copilot

Pull request overview

Copilot reviewed 14 out of 15 changed files in this pull request and generated 7 comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot · 2026-02-12T13:33:51Z

src/_compat_patches.py

+                else False,
+                warnings=warnings if warnings is not None else True,
+                mode=mode if mode is not None else "python",
+                by_alias=by_alias if by_alias is not None else True,


by_alias is forced to True when None, which changes pydantic’s default behavior (pydantic’s model_dump(by_alias=...) default is typically False). This patch may fix the TypeError but can subtly change serialized payloads sent to the Anthropics API. Prefer converting None to the original default (likely False) or dynamically deriving the default from the wrapped function/signature so the patch doesn’t alter output semantics.

Suggested change

by_alias=by_alias if by_alias is not None else True,

by_alias=by_alias if by_alias is not None else False,

Copilot · 2026-02-12T13:33:52Z

src/models/_adapter.py

+        content = self.llm.invoke(text, stop=_stop, **kwargs)
        return AIMessage(content=content)


.invoke() can return different types depending on whether self.llm is an LLM or a chat model (often str vs BaseMessage). Wrapping the return value directly into AIMessage(content=...) will produce incorrect content if the result is already a message object (you’d end up with content=<AIMessage ...>). Normalize the result (e.g., use result.content when available, otherwise use result) before constructing AIMessage.

Copilot · 2026-02-12T13:33:52Z

src/agents/base.py

-        import langchain.agents
-
-        langchain.debug = self.enable_debug_output
+        from langchain_classic.agents import AgentExecutor


self.enable_debug_output no longer influences LangChain debug output after removing langchain.debug = .... If this flag is still intended to work, reintroduce the debug toggle using the current recommended API (e.g., LangChain globals debug setter) or an equivalent mechanism compatible with the new package split.

Copilot · 2026-02-12T13:33:52Z

src/agents/base.py

        agent = agent_obj.create_agent(ctx, tools)

-        agent_exec = langchain.agents.AgentExecutor(
+        agent_exec = AgentExecutor(


self.enable_debug_output no longer influences LangChain debug output after removing langchain.debug = .... If this flag is still intended to work, reintroduce the debug toggle using the current recommended API (e.g., LangChain globals debug setter) or an equivalent mechanism compatible with the new package split.

Copilot · 2026-02-12T13:33:52Z

src/models/google/_port_types.py

            max_retries=2,  # default is 6, instead we just try twice before failing
            base_url=self.custom_base_api_url or self.base_api_url,


base_api_url now includes the scheme (https://), but custom_base_api_url (by its name) could plausibly be provided as either a host-only value or a full URL. With base_url=self.custom_base_api_url or self.base_api_url, a host-only custom value would become an invalid URL, while a full URL remains fine. Consider normalizing custom_base_api_url (e.g., add https:// if missing) or documenting/enforcing that custom_base_api_url must include the scheme to avoid ambiguous/fragile configuration.

Copilot · 2026-02-12T13:33:53Z

src/models/google/_port_types.py

+        class ChatVertexAI(ChatGoogleGenerativeAI):
+            def with_structured_output(self, *args, method="function_calling", **kwargs):
+                return super().with_structured_output(*args, method=method, **kwargs)

        google_credentials = self._construct_google_credentials()

        return ChatVertexAI(


The local subclass is named ChatVertexAI but now inherits from ChatGoogleGenerativeAI, which is confusing given Vertex vs GenAI naming. Consider renaming to something like ChatVertexAICompat/VertexAIChatCompat to better communicate that this is a compatibility shim rather than the original VertexAI class.

Suggested change

class ChatVertexAI(ChatGoogleGenerativeAI):

def with_structured_output(self, *args, method="function_calling", **kwargs):

return super().with_structured_output(*args, method=method, **kwargs)

google_credentials = self._construct_google_credentials()

return ChatVertexAI(

class ChatVertexAICompat(ChatGoogleGenerativeAI):

def with_structured_output(self, *args, method="function_calling", **kwargs):

return super().with_structured_output(*args, method=method, **kwargs)

google_credentials = self._construct_google_credentials()

return ChatVertexAICompat(

Copilot · 2026-02-12T13:33:53Z

src/indexes/chroma.py

@@ -202,43 +202,39 @@ def save_vectorstore(self, vectorstore_folder: str, vectorstore):
        from langchain_chroma import Chroma

        vectorstore: Chroma = vectorstore


The previous guard that avoided running the persist-directory workaround unless necessary was removed. If the subsequent hack reinitializes clients or reconfigures persistence, this will now happen on every save, potentially adding overhead. Consider restoring a cheap check (where possible) to skip the workaround when the vectorstore is already configured for the target folder.

Suggested change

vectorstore: Chroma = vectorstore

vectorstore: Chroma = vectorstore

# If the vectorstore is already configured to persist to the target folder,

# skip the workaround to avoid unnecessary reinitialization.

current_persist_dir = getattr(vectorstore, "persist_directory", None)

if current_persist_dir and current_persist_dir == vectorstore_folder:

return

Which now provides a unified API for Google AI Studio and Vertex AI

For backwards compatibility

Because chroma no longer provides the information where the vectorstore is persisted.

Tested locally on Windows.

AP-25545 (Update to latest langchain 1.x version)

AP-25561 ()

…r structured outputs

marc-lehner requested a review from a team as a code owner February 9, 2026 11:00

marc-lehner requested review from Copilot and knime-ghub-bot and removed request for a team February 9, 2026 11:00

Copilot AI reviewed Feb 9, 2026

View reviewed changes

marc-lehner requested a review from AtR1an February 9, 2026 13:57

Copilot AI review requested due to automatic review settings February 11, 2026 10:47

AtR1an force-pushed the enh/AP-25561-update-bundled-env-to-py-3-13 branch from 192d13c to 7bf0029 Compare February 11, 2026 10:47

Copilot AI reviewed Feb 11, 2026

View reviewed changes

Copilot AI review requested due to automatic review settings February 12, 2026 13:31

tonqui force-pushed the enh/AP-25561-update-bundled-env-to-py-3-13 branch from c5451ae to 0e38d29 Compare February 12, 2026 13:31

Copilot AI reviewed Feb 12, 2026

View reviewed changes

AtR1an and others added 18 commits February 12, 2026 14:34

AP-25545: Defer expensive imports

ff53ec1

AP-25545: Update StructuredTool import

3bfae17

AP-25545: Migrate Vertex nodes to langchain_google_genai

3712fdc

Which now provides a unified API for Google AI Studio and Vertex AI

AP-25545: Use RETRIEVAL_QUERY task type in Vertex AI Embeddings

fc546b4

For backwards compatibility

AP-25545: Always copy chroma vector store

71d0379

Because chroma no longer provides the information where the vectorstore is persisted.

AP-25545: Change imports of deprecated agent nodes to langchain_classic

067750d

AP-25545: Restore tool error behavior

5042f26

AP-25545: Update dependencies

c1382d5

AP-25545: Adapt LLMChatModelAdapter to langchain 1.x API

0437fe8

AP-25545: Monkey-patch giskard to support newer scipy

eb6cf70

AP-25545: Update and document dependencies

06fa782

Tested locally on Windows.

AP-25545: update to extension bundling 5.10 (includes pixi 0.63.2)

915bc79

AP-25545 (Update to latest langchain 1.x version)

AP-25545: add knime-python-versions 5.10 and knime-extension 5.10

19bb381

AP-25545 (Update to latest langchain 1.x version)

AP-25545: update to 5.11

94679a0

AP-25545 (Update to latest langchain 1.x version)

AP-25561: remove new ibm dependencies and update build env

8562c19

AP-25561 ()

AP-25561: Fix rebase problems

b9eaf84

AP-25561: workaround for Anthropic SDK bug

a699ba7

AP-25561: add a VertexAI wrapper that uses function_calling method fo…

7e71aa0

…r structured outputs

tonqui force-pushed the enh/AP-25561-update-bundled-env-to-py-3-13 branch from 0e38d29 to 7e71aa0 Compare February 12, 2026 13:34

		if self.user_message:
		messages.append({"role": "user", "content": self.user_message})

		@@ -202,43 +202,39 @@ def save_vectorstore(self, vectorstore_folder: str, vectorstore):
		from langchain_chroma import Chroma

+        # If the vectorstore already persists to the desired directory, skip the expensive rebuild.
+        existing_persist_dir = getattr(vectorstore, "persist_directory", None)
+        if existing_persist_dir is not None and existing_persist_dir == vectorstore_folder:
+            return

	langchain-chroma = "==1.1.0" # Direct: used in indexes/chroma.py installed via pypi because conda packages don't support python 3.13
	# Installed via PyPI because conda packages don't support Python 3.13
	langchain-chroma = "==1.1.0" # Direct: used in indexes/chroma.py

	by_alias=by_alias if by_alias is not None else True,
	by_alias=by_alias if by_alias is not None else False,

		content = self.llm.invoke(text, stop=_stop, **kwargs)
		return AIMessage(content=content)

		max_retries=2, # default is 6, instead we just try twice before failing
		base_url=self.custom_base_api_url or self.base_api_url,

		@@ -202,43 +202,39 @@ def save_vectorstore(self, vectorstore_folder: str, vectorstore):
		from langchain_chroma import Chroma

		vectorstore: Chroma = vectorstore

-        vectorstore: Chroma = vectorstore
+        vectorstore: Chroma = vectorstore
+        # If the vectorstore is already configured to persist to the target folder,
+        # skip the workaround to avoid unnecessary reinitialization.
+        current_persist_dir = getattr(vectorstore, "persist_directory", None)
+        if current_persist_dir and current_persist_dir == vectorstore_folder:
+            return

Enh/ap 25561 update bundled env to py 3 13 #42

Are you sure you want to change the base?

Enh/ap 25561 update bundled env to py 3 13 #42

Uh oh!

Conversation

marc-lehner commented Feb 9, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 9, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 11, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI Feb 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 12, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 12, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants