langchain_google_vertexai.utils create_context_cache not working , create_context_cache passes GAPIC ToolConfig, but CachedContent.create requires SDK vertexai.generative_models.ToolConfig

Summary

langchain_google_vertexai.utils.create_context_cache(...) accepts a dict-like tool_config and (per docs) converts it for you. However, it ultimately calls the Vertex SDK vertexai.preview.caching.CachedContent.create(...), which expects an SDK vertexai.generative_models.ToolConfig instance. Passing a dict (converted to GAPIC ToolConfig) triggers:

ValueError: tool_config must be a ToolConfig object

This appears to be a type mismatch between what the LangChain helper builds and what the Vertex SDK context-cache API requires.

Why this looks like a bug

LangChain’s API reference for create_context_cache documents tool_config as an internal dict (_ToolConfigDict) that the helper accepts and converts.

Vertex AI’s SDK docs show CachedContent.create(...) and function calling expect SDK types (vertexai.generative_models.ToolConfig, with FunctionCallingConfig).

Actual behavior

create_context_cache(...) raises tool_config must be a ToolConfig object when tool_config is provided as a dict (even though the helper API advertises accepting a dict and doing conversion).

Expected behavior

create_context_cache(...) should convert the dict into an SDK vertexai.generative_models.ToolConfig before calling vertexai.preview.caching.CachedContent.create(...), or otherwise route through an API that accepts a GAPIC ToolConfig. In other words, the helper should “just work” with the documented dict input.

Minimal repro

from langchain_google_vertexai import ChatVertexAI
from langchain_google_vertexai.utils import create_context_cache
from langchain_core.messages import SystemMessage

chat = ChatVertexAI(model="gemini-1.5-pro", project="...", location="...")

messages = [SystemMessage(content="Long system instructions here")]

tools = [{"function": {"name": "my_tool", "parameters": {"type": "object", "properties": {}}}}]
tool_config = {
    "function_calling_config": {
        "mode": 1,  # ANY
        "allowed_function_names": ["my_tool"]
    }
}

cache_name = create_context_cache(
    model=chat,
    messages=messages,
    tools=tools,
    tool_config=tool_config,
)