[BUG] Empty group_id not handled consistently in full-text and similarity search

[galshubeli]

Bug Description

When searching without specifying a group_id, the current implementation behaves inconsistently between full-text search and similarity search due to different filtering logic:

• Full-text search: Applies a WHERE clause using $group_ids, even when it is None, which results in no matches returned.

WHERE e.group_id IN $group_ids

• Similarity search: Does not filter by group_id when group_ids is None, and instead defaults to extracting relationships across all groups:

WHERE e.group_id IS NOT NULL

Full-text:

    query = (
        get_relationships_query('edge_name_and_fact', provider=driver.provider)
        + """
        YIELD relationship AS rel, score
        MATCH (n:Entity)-[e:RELATES_TO {uuid: rel.uuid}]->(m:Entity)
        WHERE e.group_id IN $group_ids """
        + filter_query
        + """
        WITH e, score, n, m
        RETURN
        """
        + ENTITY_EDGE_RETURN
        + """
        ORDER BY score DESC
        LIMIT $limit
        """
    )

Similarity:

    group_filter_query: LiteralString = 'WHERE e.group_id IS NOT NULL'
    if group_ids is not None:
        group_filter_query += '\nAND e.group_id IN $group_ids'
        query_params['group_ids'] = group_ids

        if source_node_uuid is not None:
            query_params['source_uuid'] = source_node_uuid
            group_filter_query += '\nAND (n.uuid = $source_uuid)'

        if target_node_uuid is not None:
            query_params['target_uuid'] = target_node_uuid
            group_filter_query += '\nAND (m.uuid = $target_uuid)'

    query = (
        RUNTIME_QUERY
        + """
        MATCH (n:Entity)-[e:RELATES_TO]->(m:Entity)
        """
        + group_filter_query
        + filter_query
        + """
        WITH DISTINCT e, n, m, """
        + get_vector_cosine_func_query('e.fact_embedding', '$search_vector', driver.provider)
        + """ AS score
        WHERE score > $min_score
        RETURN
        """
        + ENTITY_EDGE_RETURN
        + """
        ORDER BY score DESC
        LIMIT $limit
        """
    )

Steps to Reproduce

Run the quick-start examples (Falkordb/Neo4j)

Expected Behavior

Consistent group_id filtering

Actual Behavior

Inconsistent group_id filtering

Environment

• Graphiti Version: [0.18.2]
• Python Version: [3.12]
• Operating System: [Ubuntu 22.04]
• Database Backend: [Neo4j 5.26, FalkorDB 1.1.2]
• LLM Provider & Model: [OpenAI gpt-4.1]

Installation Method

• pip install
• uv add
• Development installation (git clone)

Error Messages/Traceback

Paste the full error message and traceback here

Configuration

# Relevant configuration or initialization code

Additional Context

• Does this happen consistently or intermittently?
• Which component are you using? (core library, REST server, MCP server)
• Any recent changes to your environment?
• Related issues or similar problems you've encountered?

Possible Solution

If you have ideas about what might be causing the issue or how to fix it, please share them here.

[galshubeli]

@prasmussen15 What is the expected behavior when group_id is not specified?
Should we retrieve results across all group_ids, or only from the default one?

[onestardao]

This is essentially a query-construction consistency bug — matches exactly ProblemMap No.11 in our diagnostic list (query scope divergence).

Full-text search: always applies WHERE group_id = $group_id even if $group_id is None, which results in no matches.

Similarity search: skips the filter if group_id is None and queries across all groups.

To fix:

Pick one canonical None-handling policy and apply it to both paths.

Build the WHERE clause only after explicitly checking for None, instead of passing it blindly to Cypher.

Add a unit test with group_id=None to verify identical scope in both full-text and similarity search.

If you want the full breakdown of ProblemMap No.11, I can share the link — it includes reproducible tests and a ready-to-apply patch pattern.