[Bug]: search() with a custom response_model raises ValidationError on SearchResultPayload (and error logging hangs)

[vkovac2]

Bug Description

cognee.search(..., retriever_specific_config={"response_model": MyModel}) fails after retrieval succeeds. The retriever
returns a Pydantic MyModel instance, but SearchResultPayload.completion is typed Optional[Union[str, List[str], List[dict]]], which rejects a model instance → ValidationError when the payload is built at
cognee/modules/search/methods/get_retriever_output.py:80.

Two problems:

1. The response_model search path can't return a structured object — the payload type doesn't accept a BaseModel.
2. When that ValidationError is logged, cognee's structlog → rich exception renderer recurses in rich/pretty._traverse
   and hangs indefinitely, so instead of a fast error the call appears to deadlock.

Steps to Reproduce

Minimal (no graph/LLM needed — it's pure result packaging):

from cognee.modules.search.models.SearchResultPayload import SearchResultPayload
from cognee.modules.search.types.SearchType import SearchType
from pydantic import BaseModel

class DealBrief(BaseModel):
    deal_name: str = ""
    health: str = ""

SearchResultPayload(completion=DealBrief(deal_name="Acme"), search_type=SearchType.GRAPH_COMPLETION)

End-to-end:

await cognee.search(
    query_text="Brief for the Acme deal",
    query_type=SearchType.GRAPH_COMPLETION,
    datasets="my_dataset",
    retriever_specific_config={"response_model": DealBrief},
)

Expected Behavior

search() with a response_model returns the populated model instance (schema-conforming structured output). The completion
retrievers already accept response_model via retriever_specific_config, and generate_completion returns a model
instance — only the result payload can't carry it.

Actual Behavior

pydantic ValidationError: N validation errors for SearchResultPayload at get_retriever_output.py:80:

completion.str: Input should be a valid string [input_type=DealBrief]
completion.list[str].0 / list[dict[any,any]].0: ...

Separately, logging that exception hangs indefinitely in rich/pretty._traverse (cognee/shared/logging_utils.py →
structlog → rich traceback), so the call never returns and looks like a deadlock.

Environment

Environment

• OS: macOS (Darwin 25.3.0, arm64)
• Python: 3.12.13
• Cognee: 1.1.2 (installed from dev)
• LLM Provider: Baseten (gpt-oss-120b) + OpenAI embeddings
• Database: default local Kuzu/Ladybug graph + LanceDB
• Note: the failure is store-independent — it's in post-retrieval result packaging, so it reproduces on any backend.

Logs/Error Messages

Additional Context

Root cause: SearchResultPayload.completion (in cognee/modules/search/models/SearchResultPayload.py) is
Optional[Union[str, List[str], List[dict]]] — no BaseModel, and no serializer (whereas result_object has a
field_serializer for complex types).

Suggested fix: widen completion to accept a model (e.g. add BaseModel to the Union) and model_dump() it on output;
and harden the exception logger so rich traceback rendering can't hang on the error object.

Pre-submission Checklist

• I have searched existing issues to ensure this bug hasn't been reported already
• I have provided a clear and detailed description of the bug
• I have included steps to reproduce the issue
• I have included my environment details

[Vasilije1990]

Fixed