[BUG] Crew silently continues when knowledge_sources fail to initialize (missing OPENAI_API_KEY)

[kevincostner17]

Description

When a Crew is constructed with explicit knowledge_sources, knowledge initialization can fail (for example missing OPENAI_API_KEY for the default Chroma OpenAI embedder). The failure is logged, but create_crew_knowledge catches the exception and the crew continues as if knowledge were configured.

Result: kickoff “succeeds” with empty knowledge, and agents answer without retrieval. Easy to miss in non-verbose runs.

Also: ChromaDBConfig._default_embedding_function docstring says the default is all-MiniLM-L6-v2 via ONNX, but the implementation uses OpenAIEmbeddingFunction / text-embedding-3-small (docs elsewhere correctly say OpenAI is the default).

Steps to Reproduce

from crewai import Agent, Crew, Task, Process
from crewai.knowledge.source.string_knowledge_source import StringKnowledgeSource
from crewai.llms.base_llm import BaseLLM

class StubLLM(BaseLLM):
    def call(self, messages, tools=None, callbacks=None, available_functions=None,
             from_task=None, from_agent=None, response_model=None):
        return "Thought: ok.\nFinal Answer: invented answer without retrieval."
    def supports_function_calling(self) -> bool:
        return False

# Ensure OPENAI_API_KEY is unset
knowledge = StringKnowledgeSource(content="Secret fact: the vault code is 42.")
agent = Agent(role="r", goal="g", backstory="b", llm=StubLLM(model="stub"), verbose=False)
task = Task(description="What is the vault code?", expected_output="the code", agent=agent)
crew = Crew(
    agents=[agent],
    tasks=[task],
    knowledge_sources=[knowledge],
    process=Process.sequential,
    verbose=False,
)
result = crew.kickoff()
print(result)  # completes; knowledge never loaded

Expected behavior

If the user explicitly provided knowledge_sources and initialization/upsert fails, Crew construction or kickoff should fail loudly (raise) so the misconfiguration is obvious. Optionally fix the misleading ONNX docstring on _default_embedding_function.

Actual behavior

Errors like Failed to upsert documents: The OPENAI_API_KEY environment variable is not set. are logged, then create_crew_knowledge swallows the exception and the crew runs without knowledge.

Operating System

macOS

Python Version

3.13

crewAI Version

1.15.21 / current main

Willingness to Contribute

Yes — I can open a PR that re-raises on explicit knowledge_sources init failure + corrects the Chroma default-embedder docstring, with a regression test.

[kevincostner17]

I'd like to take this. Plan:

1. Re-raise when explicit knowledge_sources fail during crew/agent knowledge init (stop silent empty-knowledge runs).
2. Correct the ChromaDBConfig._default_embedding_function docstring to match the OpenAI default.
3. Add a regression test covering missing OPENAI_API_KEY + explicit knowledge sources.

Please assign me if that sounds good — PR incoming shortly.