How can I search the best fit documents?

[LStromann]

=```python
PDF_DIRECTORY="./data"

# https://docs.cognee.ai/core-concepts/main-operations/add
# Flexible input: raw text, local files, directories, or S3 URIs
await cognee.add(PDF_DIRECTORY)

# Generate the knowledge graph
await cognee.cognify()

# Find best matched pdf between 4 pdfs
search_result = await cognee.search(
    query_text="coffee related item",
    query_type=SearchType.CHUNKS,
    top_k=3,
)

for i, result in enumerate(search_result):
    print(f"result {i}: {type(result)}")
    print(result)

As a result, I get a dictionary like the following:
`{'id': 'cec8bdff-b615-5cc2-9612-0cd0d7052b20', 'created_at': 1765802162917, 'updated_at': 1765802162917, 'ontology_valid': False, 'version': 1, 'topological_rank': 0, 'type': 'IndexSchema', 'text': '...'}` 
However, I can’t find a way to retrieve or construct file-level metadata (such as the PDF filename) from a chunk.
Is there a recommended way to associate or recover the original file metadata from chunks?

Any help would be appreciated.

This discussion was automatically pulled from [Discord](https://discord.gg/m63hxKsp4p).

[LStromann]

hey <@315485233287790602> what version of cognee are you running?

also can you print what fields you're actually getting?

for result in search_result: print(result.keys())

if you don't see document info in there, try using the get dataset data endpoint at https://docs.cognee.ai/api-reference/datasets/get-dataset-data - it shows all files with their names and IDs.

or try SearchType.INSIGHTS instead of SearchType.CHUNKS - might preserve document-level info better.

also you can check the blog here might help a bit https://dev.to/chinmay_bhosale_9ceed796b/search-types-in-cognee-1jo7

[LStromann]

<@1373135538147692604>
thanks for your reply 😃

• Version I'm using is 0.4.1
• fields actually returned are ['id', 'created_at', 'updated_at', 'ontology_valid', 'version', 'topological_rank', 'type', 'text']

I find that I can get data by API /api/v1/datasets/{my_dataset_id}/data , but still clue-less to how can I map result['id'] to data['id'], as you mentioned.
However, I’m still unclear on how to map result['id'] to data['id'].
For example, when I query the API, I receive the following list of data:

[
  {
    "id": "4449f767-ef91-5b12-803a-580f003aefad",
    "name": "1020250180918A",
    ...
    "datasetId": "40e38dea-1cf0-5ddf-9637-f3c15efe87b5"
  },
  {
    "id": "f5213b99-d7f9-5876-8130-1c08ce2c9a77",
    "name": "1020220034532B1",
    ...
    "datasetId": "40e38dea-1cf0-5ddf-9637-f3c15efe87b5"
  },
  {
    "id": "69d53292-3886-565a-98f2-68365e1c8a47",
    "name": "2020240001358Y1",
    ...
    "datasetId": "40e38dea-1cf0-5ddf-9637-f3c15efe87b5"
  },
  {
    "id": "fec20126-646b-57e3-b866-8690ada72319",
    "name": "1020230051009B1",
    ...
    "datasetId": "40e38dea-1cf0-5ddf-9637-f3c15efe87b5"
  }
]

On the other hand, the search result looks like this:
{'id': 'cec8bdff-b615-5cc2-9612-0cd0d7052b20', 'created_at': 1765802162917, 'updated_at': 1765802162917, 'ontology_valid': False, 'version': 1, 'topological_rank': 0, 'type': 'IndexSchema', 'text': 'some_texts'}

How can I map this search result back to the corresponding file-level metadata (such as the file ID or name)?

[LStromann]

hey <@315485233287790602> if your goal is to find the best matching pdf overall not just individual chunks I'd suggest using SearchType.GRAPH_COMPLETION instead.

from the discussions here https://discord.com/channels/1120795297094832337/1356574585398759467 <@508682252595626015> recommends it for most general questions since it does graph traversal and gives you better context.

[LStromann]

<@1373135538147692604> thanks. it looks like exactly I was looking for 🙂

After I've upgraded cognee to v0.5.0, when I curl to local cognee server, I get {"detail":"Unauthorized"} response.

2025-12-16T05:13:23.317806 [warning  ] From version 0.5.0 onwards, Cognee will run with multi-user access control mode set to on by default. Data isolation between different users and datasets will be enforced and data created before multi-user access control mode was turned on won't be accessible by default. To disable multi-user access control mode and regain access to old data set the environment variable ENABLE_BACKEND_ACCESS_CONTROL to false before starting Cognee. For more information, please refer to the Cognee documentation. [cognee.shared.logging_utils]

[LStromann]

I found this warning but can't figure out where should I put ENABLE_BACKEND_ACCESS_CONTROL=false to send curl request without any auth

[LStromann]

nevermind, I found a way in https://docs.cognee.ai/setup-configuration/permissions. nice docs!

[LStromann]

with SearchType.GRAPH_COMPLETION , I can get the results filled with some graph info as below.

{'dataset_id': UUID('af95cb27-101b-5d59-b256-248b67e570e9'),
 'dataset_name': 'main_dataset',
 'dataset_tenant_id': None,
 'graphs': {'main_dataset': {'edges': [{'label': 'contains',
                                        'source': '7b2b8913-...',
                                        'target': '9f96e93f-...'},
                                       {'label': 'contains',
                                        'source': '330d1e94-...',
                                        'target': '0e2ff7fa-...'},
                                       {'label': 'contains',
                                        'source': '7b2b8913-...',
                                        'target': 'd37496d0-...'}],
                             'nodes': [{'attributes': {'description': None,
                                                       'id': None,
                                                       'name': '',
                                                       'text': '...\n',
                                                       'type': 'DocumentChunk',
                                                       'vector_distance': 0.0931496685487514},
                                        'id': '7b2b8913-...',
                                        'label': '',
                                        'type': 'DocumentChunk'},

However, none of these uids are pointing document object directly. 🥹
Are there any way to map search result and data file?

[                                                
  {                                              
    "id": "1d750d4a-...",
    "name": "1020250180918A",                    
    "createdAt": "2025-12-16T05:08:41.986599",   
    "updatedAt": "2025-12-16T05:10:50.467125",   
    "extension": "txt",                          
    "mimeType": "text/plain",                    

[LStromann]

can you try this

search_result = await cognee.search(
    query_text="coffee related item",
    query_type=SearchType.GRAPH_COMPLETION,
)

graph_engine = await get_graph_engine()

# extract chunk IDs from the graph results
for dataset_results in search_result:
    graphs = dataset_results['graphs']

    for dataset_name, graph_data in graphs.items():
        print(f"\nDataset: {dataset_name}")

        chunk_ids = [node['id'] for node in graph_data['nodes']
                     if node['type'] == 'DocumentChunk']

        # for each chunk, find its parent document
        for chunk_id in chunk_ids:
            connections = await graph_engine.get_connections(chunk_id)

            for source, edge, target in connections:
                if edge.get('relationship_name') == 'is_part_of':
                    print(f"Chunk from document: {target['name']} (ID: {target['id']})")
                    break

this uses the graph store structural search capabilities to navigate from chunks to documents. lmk if this helps

[LStromann]

Thanks for the snippet.
I've run the code you gave and found that all connections are empty 🤔

            for chunk_id in chunk_ids:
                connections = await graph_engine.get_connections(chunk_id)
                print(f"connections: {connections}") # -> connections: []

[LStromann]

After add memify after cognify, connections are created and finally can find document name!
Thanks for your support! 😄