INDEXING.md

Indexing — Agentic GraphRAG

MongoDB indexes for the graphrag database. Run after any fresh init() to ensure all performance indexes are in place.

Collections & Indexes

entities

Index	Key	Purpose
entity_id_1	{entity_id: 1}	Direct lookup by entity_id
name_lower_1	{name_lower: 1}	Case-insensitive entity lookup (unique)
canonical_id_1	{canonical_id: 1}	Alias resolution (follow canonical chain)
entity_type_1	{entity_type: 1}	Filter entities by type
name_text	text index on name	Full-text entity search
entity_type_name_lower	{entity_type, name_lower}	Type-filtered + name lookup
canonical_id_entity_type	{canonical_id, entity_type}	Canonical chain + type filtering

relations

Index	Key	Purpose
relation_id_1	{relation_id: 1}	Direct lookup by relation_id
subject_id_1	{subject_id: 1}	Graph traversal from subject
object_id_1	{object_id: 1}	Graph traversal from object
predicate_1	{predicate: 1}	Filter by relation type
subject_id_1_predicate_1	{subject_id, predicate}	Traversal + predicate filter
object_id_1_predicate_1	{object_id, predicate}	Traversal + predicate filter
predicate_created_at	{predicate, created_at}	Temporal queries by type
subject_predicate_object	{subject_id, predicate, object_id}	Full triplet lookup

documents

Index	Key	Purpose
text_index	text index on content	Full-text document search
doc_id_1	{doc_id: 1}	Direct document lookup (unique)
source_1	{source: 1}	Filter by source (url, file, etc.)
created_at_1	{created_at: 1}	Temporal queries
source_created_at	{source, created_at}	Source-scoped temporal queries

Creating Indexes

cd /home/doug/Agentic\ GraphRAG
source .venv/bin/activate
python3 -c "
from src import mongo_memory
mongo_memory.init()

from pymongo import ASCENDING
db = mongo_memory._get_db()

indexes = [
    (mongo_memory.ENTITIES_COL, [('entity_type', ASCENDING), ('name_lower', ASCENDING)], 'entity_type_name_lower'),
    (mongo_memory.RELATIONS_COL, [('predicate', ASCENDING), ('created_at', ASCENDING)], 'predicate_created_at'),
    (mongo_memory.RELATIONS_COL, [('subject_id', ASCENDING), ('predicate', ASCENDING), ('object_id', ASCENDING)], 'subject_predicate_object'),
    (mongo_memory.ENTITIES_COL, [('canonical_id', ASCENDING), ('entity_type', ASCENDING)], 'canonical_id_entity_type'),
    (mongo_memory.DOCUMENTS_COL, [('source', ASCENDING), ('created_at', ASCENDING)], 'source_created_at'),
]

for col, fields, name in indexes:
    try:
        db[col].create_index(fields, name=name)
        print(f'Created: {col}.{name}')
    except Exception as e:
        print(f'Skip/Error {col}.{name}: {e}')
"

Benchmarking

source .venv/bin/activate
python3 src/benchmark_queries.py

See benchmark_queries.py for query-level latency measurements.