supabase-integration（Supabase 集成）

Supabase Integration for Mem0 OSS

Complete guide for setting up Supabase as the backend for Mem0 Open Source (self-hosted) mode, including PostgreSQL schema with pgvector, RLS policies for security, and production-ready configurations.

Instructions

Phase 1: Supabase Project Setup

Prerequisites Check:

1. Verify Supabase is initialized: bash scripts/verify-supabase-setup.sh
2. If not initialized, set up Supabase first:

- Run /supabase:init command - Note down project ID and connection details - Obtain connection string from Supabase dashboard

Environment Configuration:

# Required environment variables
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_ANON_KEY=your-anon-key
SUPABASE_SERVICE_KEY=your-service-key
SUPABASE_DB_URL=postgresql://postgres:[password]@db.[project].supabase.co:5432/postgres

# Optional: Mem0-specific configs
MEM0_EMBEDDING_MODEL=text-embedding-3-small
MEM0_VECTOR_DIMENSION=1536

Phase 2: Enable pgvector Extension

Enable Extension:

bash scripts/setup-mem0-pgvector.sh

This script:

1. Connects to Supabase database
2. Enables pgvector extension
3. Verifies extension is active
4. Checks PostgreSQL version compatibility (>= 12)

Manual Verification:

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

-- Test vector operations
SELECT '[1,2,3]'::vector;

Phase 3: Create Memory Tables Schema

Apply Memory Schema:

bash scripts/apply-mem0-schema.sh

This creates three core tables:

1. memories table (vector storage):

• id (uuid, primary key)
• user_id (text, indexed) - User identifier for isolation
• agent_id (text, indexed, nullable) - Agent identifier
• run_id (text, indexed, nullable) - Session/conversation identifier
• memory (text) - Memory content
• hash (text) - Content hash for deduplication
• metadata (jsonb) - Flexible metadata storage
• categories (text[]) - Memory categorization
• embedding (vector(1536)) - Semantic embedding
• created_at (timestamptz)
• updated_at (timestamptz)

2. memory_relationships table (graph memory):

• id (uuid, primary key)
• source_memory_id (uuid, foreign key)
• target_memory_id (uuid, foreign key)
• relationship_type (text) - e.g., "references", "caused_by", "related_to"
• strength (numeric) - Relationship strength (0.0-1.0)
• metadata (jsonb)
• user_id (text, indexed) - For RLS isolation
• created_at (timestamptz)

3. memory_history table (audit trail):

• id (uuid, primary key)
• memory_id (uuid)
• operation (text) - "create", "update", "delete"
• old_value (jsonb)
• new_value (jsonb)
• user_id (text)
• timestamp (timestamptz)

Use Template for Custom Schema:

# Generate schema with custom dimensions
bash scripts/generate-mem0-schema.sh \
  --dimensions 1536 \
  --include-graph true \
  --include-history true \
  > custom-mem0-schema.sql

# Apply custom schema
psql $SUPABASE_DB_URL < custom-mem0-schema.sql

Phase 4: Create Performance Indexes

Apply Optimized Indexes:

bash scripts/create-mem0-indexes.sh

Index Strategy:

1. Vector Search Indexes (HNSW): -- Main embedding index (cosine distance) CREATE INDEX idx_memories_embedding ON memories USING hnsw (embedding vector_cosine_ops) WITH (m = 16, ef_construction = 64);
2. User/Agent Isolation Indexes: CREATE INDEX idx_memories_user_id ON memories(user_id); CREATE INDEX idx_memories_agent_id ON memories(agent_id); CREATE INDEX idx_memories_run_id ON memories(run_id);
3. Composite Indexes for Common Queries: -- User + timestamp for chronological retrieval CREATE INDEX idx_memories_user_created ON memories(user_id, created_at DESC); -- User + agent for agent-specific memories CREATE INDEX idx_memories_user_agent ON memories(user_id, agent_id);
4. Graph Relationship Indexes: CREATE INDEX idx_relationships_source ON memory_relationships(source_memory_id); CREATE INDEX idx_relationships_target ON memory_relationships(target_memory_id); CREATE INDEX idx_relationships_type ON memory_relationships(relationship_type);
5. Full-Text Search Index (optional): CREATE INDEX idx_memories_content_fts ON memories USING gin(to_tsvector('english', memory));

Index Selection Guide:

• Small dataset (< 100K memories): Start with basic indexes
• Medium dataset (100K-1M): Add HNSW with m=16
• Large dataset (> 1M): Use HNSW with m=32, consider IVFFlat
• Write-heavy workload: Consider IVFFlat over HNSW

Phase 5: Implement Row Level Security (RLS)

Apply RLS Policies:

bash scripts/apply-mem0-rls.sh

Security Patterns:

1. User Isolation (Default):

-- Users can only access their own memories
CREATE POLICY "Users access own memories"
ON memories FOR ALL
USING (auth.uid()::text = user_id);

-- Users can only see their own relationships
CREATE POLICY "Users access own relationships"
ON memory_relationships FOR ALL
USING (auth.uid()::text = user_id);

2. Multi-Tenant Isolation (Enterprise):

-- Check organization membership
CREATE POLICY "Organization members access memories"
ON memories FOR ALL
USING (
  EXISTS (
    SELECT 1 FROM org_members
    WHERE org_members.user_id = auth.uid()::text
    AND org_members.org_id = memories.metadata->>'org_id'
  )
);

3. Agent-Specific Policies:

-- Public agent memories (shared across users)
CREATE POLICY "Agent memories readable by all"
ON memories FOR SELECT
USING (agent_id IS NOT NULL AND user_id IS NULL);

-- Agent can write to their own memory space
CREATE POLICY "Agent writes own memories"
ON memories FOR INSERT
WITH CHECK (agent_id = current_setting('app.agent_id', true));

Test RLS Enforcement:

bash scripts/test-mem0-rls.sh --user-id "test-user-123"

Phase 6: Configure Connection Pooling

Setup PgBouncer (Recommended for Production):

bash scripts/configure-connection-pool.sh

Connection Pooling Strategy:

For Mem0 OSS:

• Transaction mode (default): Each memory operation gets fresh connection
• Session mode: Use for graph traversals requiring multiple queries
• Pool size: Start with 20 connections, scale based on load

Configuration:

# Mem0 with connection pooling
from mem0 import Memory

config = {
    "vector_store": {
        "provider": "postgres"
        "config": {
            "url": "postgresql://user:pass@pooler.project.supabase.co:6543/postgres"
            "pool_size": 20
            "max_overflow": 10
            "pool_timeout": 30
            "pool_recycle": 3600
        }
    }
}

memory = Memory.from_config(config)

Supabase Pooler URLs:

• Transaction mode: pooler.project.supabase.co:6543
• Session mode: pooler.project.supabase.co:5432

Phase 7: Implement Backup Strategy

Setup Automated Backups:

bash scripts/setup-mem0-backup.sh --schedule daily --retention 30

Backup Strategies:

1. Point-in-Time Recovery (Supabase Built-in):

• Automatic backups (Pro plan and above)
• Restore to any point in last 7-30 days
• No manual configuration needed

2. Manual SQL Dumps:

# Full database backup
bash scripts/backup-mem0-memories.sh

# Incremental backup (changes since last backup)
bash scripts/backup-mem0-memories.sh --incremental --since "2025-10-01"

# Backup to S3
bash scripts/backup-mem0-memories.sh --destination s3://my-bucket/mem0-backups/

3. Selective Backups:

# Backup specific user's memories
bash scripts/backup-user-memories.sh --user-id "customer-123"

# Backup by date range
bash scripts/backup-mem0-memories.sh --from "2025-01-01" --to "2025-10-27"

Restore Procedures:

# Restore from backup file
bash scripts/restore-mem0-backup.sh backup-2025-10-27.sql

# Restore specific user
bash scripts/restore-user-memories.sh backup-user-123.sql --user-id "customer-123"

Phase 8: Migration from Platform to OSS

Export from Mem0 Platform:

bash scripts/export-from-platform.sh \
  --api-key "your-platform-api-key" \
  --output platform-export.json

Transform and Import to Supabase:

bash scripts/migrate-platform-to-oss.sh \
  --input platform-export.json \
  --supabase-url $SUPABASE_URL \
  --dry-run  # Test first

# After validation, run actual migration
bash scripts/migrate-platform-to-oss.sh \
  --input platform-export.json \
  --supabase-url $SUPABASE_URL

Migration Steps:

1. Export memories from Platform API
2. Transform format (Platform JSON → Postgres schema)
3. Generate embeddings if missing
4. Validate data integrity
5. Batch insert to Supabase
6. Verify counts and sample queries
7. Update application configs to use OSS

Rollback Plan:

# Create migration checkpoint before starting
bash scripts/create-migration-checkpoint.sh

# Rollback if issues occur
bash scripts/rollback-migration.sh --checkpoint checkpoint-2025-10-27

Phase 9: Validation and Testing

Run Complete Validation Suite:

bash scripts/validate-mem0-setup.sh

Validation Checks:

• ✅ pgvector extension enabled
• ✅ All tables created with correct schema
• ✅ Indexes created and being used
• ✅ RLS policies active and enforcing
• ✅ Connection pooling configured
• ✅ Backup system operational
• ✅ Sample memory CRUD operations working
• ✅ Vector search returning results
• ✅ Graph relationships functional (if enabled)

Performance Benchmarks:

bash scripts/benchmark-mem0-performance.sh

Expected Performance (1K-10K memories):

• Memory insertion: < 50ms
• Vector search (top 10): < 100ms
• Memory retrieval by ID: < 10ms
• Graph traversal (1-2 hops): < 150ms

Configuration Templates

Template 1: Basic Mem0 OSS + Supabase

Use template: templates/mem0-basic-config.py

from mem0 import Memory

config = {
    "vector_store": {
        "provider": "postgres"
        "config": {
            "url": "postgresql://postgres:[password]@db.[project].supabase.co:5432/postgres"
            "table_name": "memories"
            "embedding_dimension": 1536
        }
    }
}

memory = Memory.from_config(config)

Template 2: Full-Featured (Vector + Graph)

Use template: templates/mem0-graph-config.py

from mem0 import Memory

config = {
    "vector_store": {
        "provider": "postgres"
        "config": {
            "url": os.getenv("SUPABASE_DB_URL")
            "table_name": "memories"
            "embedding_dimension": 1536
        }
    }
    "graph_store": {
        "provider": "postgres",  # Using same DB for graph
        "config": {
            "url": os.getenv("SUPABASE_DB_URL")
            "relationship_table": "memory_relationships"
        }
    }
    "version": "v1.1"
}

memory = Memory.from_config(config)

Template 3: Enterprise Multi-Tenant

Use template: templates/mem0-enterprise-config.py

Includes:

• Organization-level isolation
• Role-based access control
• Audit logging
• Advanced RLS policies
• Cost tracking per tenant

Common Patterns

Pattern 1: User Memory Isolation

Scenario: SaaS app with user-specific memories

Implementation: See examples/user-isolation-pattern.md

Key Points:

• Always filter by user_id
• RLS policies prevent cross-user access
• Use composite indexes (user_id + created_at)

Pattern 2: Multi-Tenant Organization

Scenario: Teams/organizations share memories within org

Implementation: See examples/multi-tenant-pattern.md

Key Points:

• Add org_id to metadata
• RLS checks org membership
• Hierarchical access (org admins see all)

Pattern 3: Agent Knowledge Base

Scenario: Shared agent memories across all users

Implementation: See examples/agent-knowledge-pattern.md

Key Points:

• agent_id not null, user_id null for shared memories
• Separate RLS policies for public agent knowledge
• Versioning for agent memory updates

Pattern 4: Session-Based Memory

Scenario: Temporary conversation context

Implementation: See examples/session-memory-pattern.md

Key Points:

• Use run_id for session identification
• Auto-cleanup after session expiry
• Promote important memories to user level

Troubleshooting

pgvector Extension Issues

Problem: Extension not available

ERROR: extension "vector" is not available

Solution:

1. Verify PostgreSQL version >= 12
2. Enable in Supabase dashboard: Database → Extensions → vector
3. Wait 1-2 minutes for activation
4. Verify: SELECT * FROM pg_extension WHERE extname = 'vector';

Slow Vector Search

Problem: Queries taking > 500ms

Solutions:

1. Check index exists: SELECT indexname FROM pg_indexes WHERE tablename = 'memories' AND indexname LIKE '%embedding%';
2. Reduce search space with filters: memory.search(query, user_id="specific-user", limit=5)
3. Increase HNSW parameters (rebuild index): DROP INDEX idx_memories_embedding; CREATE INDEX idx_memories_embedding ON memories USING hnsw (embedding vector_cosine_ops) WITH (m = 32, ef_construction = 128);
4. Switch to IVFFlat for large datasets (> 1M): bash scripts/migrate-to-ivfflat.sh

RLS Blocking Queries

Problem: Queries return empty results despite data existing

Solution:

1. Check RLS is enabled: SELECT tablename, rowsecurity FROM pg_tables WHERE tablename = 'memories';
2. Verify auth context is set: # Ensure user_id in JWT supabase.auth.get_user()
3. Test with service key (bypasses RLS): bash scripts/test-mem0-rls.sh --bypass-rls
4. Review policy logs: bash scripts/debug-rls-policies.sh

Connection Pool Exhaustion

Problem: "too many connections" errors

Solutions:

1. Use transaction pooler (port 6543)
2. Increase pool size in config
3. Implement connection retry logic
4. Monitor connection usage: bash scripts/monitor-connections.sh

Migration Failures

Problem: Migration script fails partway through

Recovery:

# Check migration status
bash scripts/check-migration-status.sh

# Rollback to checkpoint
bash scripts/rollback-migration.sh --checkpoint last

# Resume from last successful batch
bash scripts/resume-migration.sh

Security Best Practices

Checklist

• ✅ RLS enabled on all Mem0 tables
• ✅ Service key never exposed to client
• ✅ All queries filtered by user_id/org_id
• ✅ Connection strings use environment variables
• ✅ SSL/TLS enforced for database connections
• ✅ Regular security audits run
• ✅ Sensitive memory content encrypted at rest
• ✅ Backup files encrypted
• ✅ Access logs monitored for anomalies
• ✅ GDPR/data deletion policies implemented

Audit Script

Run regular security audits:

bash scripts/audit-mem0-security.sh --report security-audit.md

Checks:

• RLS policy coverage
• Unprotected tables
• Missing indexes on security columns
• Suspicious access patterns
• Cross-user query attempts
• Service key usage in logs

Files Reference

Scripts (all executable, production-ready):

• scripts/verify-supabase-setup.sh - Check Supabase initialization
• scripts/setup-mem0-pgvector.sh - Enable pgvector extension
• scripts/apply-mem0-schema.sh - Create memory tables
• scripts/generate-mem0-schema.sh - Generate custom schema
• scripts/create-mem0-indexes.sh - Create optimized indexes
• scripts/apply-mem0-rls.sh - Apply RLS policies
• scripts/test-mem0-rls.sh - Test security policies
• scripts/configure-connection-pool.sh - Setup pooling
• scripts/setup-mem0-backup.sh - Configure backups
• scripts/backup-mem0-memories.sh - Manual backup
• scripts/restore-mem0-backup.sh - Restore from backup
• scripts/backup-user-memories.sh - User-specific backup
• scripts/export-from-platform.sh - Export Platform memories
• scripts/migrate-platform-to-oss.sh - Platform → OSS migration
• scripts/validate-mem0-setup.sh - Complete validation
• scripts/benchmark-mem0-performance.sh - Performance testing
• scripts/migrate-to-ivfflat.sh - Switch to IVFFlat index
• scripts/debug-rls-policies.sh - Debug RLS issues
• scripts/monitor-connections.sh - Connection monitoring
• scripts/audit-mem0-security.sh - Security audit

Templates:

• templates/mem0-schema.sql - Base schema with pgvector
• templates/mem0-schema-graph.sql - Schema with graph support
• templates/mem0-indexes.sql - Performance indexes
• templates/mem0-rls-policies.sql - Security policies
• templates/mem0-basic-config.py - Basic Python config
• templates/mem0-graph-config.py - Full-featured config
• templates/mem0-enterprise-config.py - Multi-tenant config
• templates/backup-policy.yaml - Backup configuration
• templates/connection-pool-config.ini - PgBouncer config

Examples:

• examples/user-isolation-pattern.md - User-specific memories
• examples/multi-tenant-pattern.md - Organization isolation
• examples/agent-knowledge-pattern.md - Shared agent memories
• examples/session-memory-pattern.md - Temporary session context
• examples/platform-to-oss-migration-guide.md - Complete migration walkthrough
• examples/backup-restore-procedures.md - Disaster recovery guide
• examples/performance-tuning-guide.md - Optimization strategies

Plugin: mem0 Version: 1.0.0 Last Updated: 2025-10-27