crewai-memory-postgres

A PostgreSQL (pgvector) StorageBackend for CrewAI's unified Memory — hierarchical scopes, categories, metadata filters, importance/recency, and native vector search (<=>, HNSW index).

pip install crewai-memory-postgres

from crewai import Crew
from crewai.memory import Memory
from crewai_memory_postgres import PostgresMemoryBackend

backend = PostgresMemoryBackend("postgresql://user:pass@host:5432/db", dimensions=3072)   # match your embedder
memory = Memory(storage=backend)
crew = Crew(agents=[...], tasks=[...], memory=memory)

Or once for every Crew(memory=True): set_memory_storage_factory(lambda spec: PostgresMemoryBackend(...)).

How it works

• One table (crewai_memory by default) with id, scope, content, categories / metadata (JSONB), importance, timestamps, source, private, and embedding vector(dims); a B-tree on scope and an HNSW cosine index on embedding. Auto-created; CREATE EXTENSION IF NOT EXISTS vector is run at startup.
• search is one SQL query ordered by embedding <=> query with the scope subtree in WHERE (scope = p OR scope LIKE 'p/%'); score is 1 - cosine_distance. categories / metadata_filter / min_score are applied on the candidates.
• Requires the pgvector extension on the server (apt install postgresql-16-pgvector, or built in on RDS / Cloud SQL / Azure Database).

Docs: https://skamalj.github.io/agentstate-reducer/ · part of crewai-memory

License

MIT