t6_mem0_v2/migrations/supabase/README.md

Supabase Migrations for T6 Mem0 v2

Overview

This directory contains SQL migrations for setting up the Supabase vector store used by T6 Mem0 v2.

Migrations

001_init_vector_store.sql

Initial setup migration that creates:

• pgvector extension: Enables vector similarity search
• t6_memories table: Main storage for memory vectors and metadata
• Indexes: HNSW for vectors, B-tree for filters, GIN for JSONB
• Functions:
  • match_t6_memories(): Vector similarity search with filters
  • get_t6_memory_stats(): Memory statistics
  • update_t6_memories_updated_at(): Auto-update timestamp
• View: t6_recent_memories for quick access to recent entries

Applying Migrations

Method 1: Supabase SQL Editor (Recommended)

1. Open your Supabase project dashboard
2. Navigate to SQL Editor
3. Create a new query
4. Copy and paste the contents of 001_init_vector_store.sql
5. Click "Run" to execute

Method 2: psql Command Line

# Connect to your Supabase database
psql "postgresql://supabase_admin:PASSWORD@172.21.0.12:5432/postgres"

# Run the migration
\i migrations/supabase/001_init_vector_store.sql

Method 3: Programmatic Application

import psycopg2

# Connect to Supabase
conn = psycopg2.connect(
    "postgresql://supabase_admin:PASSWORD@172.21.0.12:5432/postgres"
)

# Read and execute migration
with open('migrations/supabase/001_init_vector_store.sql', 'r') as f:
    migration_sql = f.read()

with conn.cursor() as cur:
    cur.execute(migration_sql)
    conn.commit()

conn.close()

Verification

After applying the migration, verify the setup:

-- Check if pgvector extension is enabled
SELECT * FROM pg_extension WHERE extname = 'vector';

-- Check if table exists
\d t6_memories

-- Verify indexes
\di t6_memories*

-- Test the similarity search function
SELECT * FROM match_t6_memories(
    '[0.1, 0.2, ...]'::vector(1536),  -- Sample embedding
    10,                                -- Match count
    'test_user',                       -- User ID filter
    NULL,                              -- Agent ID filter
    NULL                               -- Run ID filter
);

-- Get memory statistics
SELECT * FROM get_t6_memory_stats();

Rollback

If you need to rollback the migration:

-- Drop view
DROP VIEW IF EXISTS t6_recent_memories;

-- Drop functions
DROP FUNCTION IF EXISTS get_t6_memory_stats();
DROP FUNCTION IF EXISTS match_t6_memories(vector, INT, TEXT, TEXT, TEXT);
DROP FUNCTION IF EXISTS update_t6_memories_updated_at();

-- Drop trigger
DROP TRIGGER IF EXISTS t6_memories_updated_at_trigger ON t6_memories;

-- Drop table (WARNING: This will delete all data!)
DROP TABLE IF EXISTS t6_memories CASCADE;

-- Optionally remove extension (only if not used elsewhere)
-- DROP EXTENSION IF EXISTS vector CASCADE;

Schema

t6_memories Table

Column	Type	Description
id	UUID	Primary key
embedding	vector(1536)	OpenAI embedding vector
metadata	JSONB	Flexible metadata
user_id	TEXT	User identifier
agent_id	TEXT	Agent identifier
run_id	TEXT	Run identifier
memory_text	TEXT	Original memory text
created_at	TIMESTAMPTZ	Creation timestamp
updated_at	TIMESTAMPTZ	Last update timestamp
hash	TEXT	Deduplication hash (unique)

Indexes

• t6_memories_embedding_idx: HNSW index for fast vector search
• t6_memories_user_id_idx: B-tree for user filtering
• t6_memories_agent_id_idx: B-tree for agent filtering
• t6_memories_run_id_idx: B-tree for run filtering
• t6_memories_created_at_idx: B-tree for time-based queries
• t6_memories_metadata_idx: GIN for JSON queries
• t6_memories_text_search_idx: GIN for full-text search

Notes

• The HNSW index provides O(log n) approximate nearest neighbor search
• Cosine distance is used for similarity (1 - cosine similarity)
• All timestamps are stored in UTC
• The hash column ensures deduplication of identical memories
• Metadata is stored as JSONB for flexible schema evolution

Support

For issues or questions about migrations, refer to:

• Supabase Vector Documentation
• pgvector Documentation
• Project Architecture: ../../ARCHITECTURE.md