Fix: Add vector store config, fix volume mount, add Ollama support

[sborenst]

Written with Cursor, works on my machine.

Pull Request: Fix Vector Store Configuration & Add Ollama Support

Summary

This PR fixes critical bugs preventing OpenMemory from working and adds comprehensive Ollama integration documentation.

Problems Fixed

1. Missing Vector Store Configuration (Critical)

Issue: Fresh installations fail with "Memory client is not available" error because vector store configuration is missing.

Root Cause: config.py line 72 returns "vector_store": None, which gets saved to the database and prevents the memory client from initializing.

Files Changed:

• openmemory/api/app/routers/config.py - Added proper Qdrant configuration
• openmemory/api/config.json - Added vector_store section for consistency
• openmemory/api/default_config.json - Added vector_store section for consistency

2. Incorrect Volume Mount

Issue: Docker Compose volume mount overwrites the entire application directory, causing "Could not import module main" errors.

Root Cause: Mounting ./api:/usr/src/openmemory replaces the Python code built into the container.

Fix: Changed to only mount the data directory: ./api/data:/usr/src/openmemory/data

Files Changed:

• openmemory/docker-compose.yml - Fixed volume mount path

3. Environment Variable Precedence

Issue: Database configuration overrides environment variables, preventing compose files from controlling Qdrant connection settings.

Root Cause: Database config always overwrote environment-based config, even when QDRANT_HOST and QDRANT_PORT were set.

Fix: Environment variables now take precedence over database configuration for vector store settings, allowing compose files to control connection settings.

Files Changed:

• openmemory/api/app/utils/memory.py - Added environment variable precedence check

Enhancements

Ollama Integration Documentation

Added comprehensive guide for switching from OpenAI to Ollama for local AI processing.

Files Added:

• openmemory/OLLAMA.md - Complete Ollama integration guide
• openmemory/podman-compose.yml - Podman-compatible compose file with SELinux support

Covers:

• Model installation
• Configuration steps
• Dimension handling (1536 vs 768)
• Testing procedures
• Troubleshooting
• Switching back to OpenAI
• Podman support

Testing

Tested Scenarios

✅ Fresh install with fixed configuration
✅ Memory creation with OpenAI
✅ Switching to Ollama
✅ Memory creation with Ollama (embeddings working)
✅ Environment variable precedence
✅ Podman compatibility
✅ Service name resolution (mem0_store)
✅ Host gateway access (host.containers.internal)
✅ Data persistence across restarts

Test Commands

# Clean install test
rm -rf api/data/
docker-compose up -d
sleep 30

# Create memory (should work without reset)
curl -X POST http://localhost:8765/api/v1/memories/ \
  -H "Content-Type: application/json" \
  -d '{"user_id": "test", "text": "Testing fixed configuration"}'

# Verify
curl http://localhost:8765/api/v1/memories/?user_id=test

Impact

Before This PR

• ❌ Fresh installations don't work
• ❌ Requires manual config reset workaround
• ❌ Volume mount causes startup failures
• ❌ No documentation for Ollama integration
• ❌ Users stuck and confused

After This PR

• ✅ Fresh installations work immediately
• ✅ No manual configuration needed
• ✅ Stable container startup
• ✅ Clear Ollama integration path
• ✅ Better user experience

Files Changed

openmemory/
├── api/
│   ├── app/
│   │   ├── routers/config.py       # Fixed vector_store default
│   │   └── utils/memory.py         # Environment variable precedence
│   ├── config.json                 # Added vector_store section
│   └── default_config.json         # Added vector_store section
├── docker-compose.yml              # Fixed volume mount, added network
├── podman-compose.yml              # New: Podman support
└── OLLAMA.md                       # New: Ollama guide

Backward Compatibility

✅ Fully backward compatible

• Existing installations continue working
• No breaking changes to API
• Configuration format unchanged
• Only adds missing required configuration

Additional Notes

Why Vector Store Config is Critical

Mem0 requires vector store configuration to initialize the memory client. Without it:

1. Memory() constructor fails
2. No memories can be created
3. API returns "Memory client is not available"
4. System is unusable

Why Volume Mount Matters

The current volume mount (./api:/usr/src/openmemory) overwrites:

• All Python dependencies installed in the container
• The application code itself
• Required system files

This causes immediate startup failures. The fix mounts only the data directory, preserving the container's built code while maintaining data persistence.

Ollama Integration Benefits

• Cost: Free vs paid OpenAI API
• Privacy: Local processing, no data sent externally
• Offline: Works without internet
• Trade-off: ~2x slower than OpenAI, but acceptable for personal use

Checklist

• Code changes tested locally
• Documentation added
• Backward compatible
• No breaking changes
• Ready for review

Known Issues

See KNOWN_ISSUES.md for details on:

• Memory categorization with Ollama (non-blocking, core functionality works)

Future Improvements (Not in This PR)

• Automated switching script
• Fix memory categorization with Ollama
• Environment variable templates
• Additional LLM provider documentation

[CLAassistant]

All committers have signed the CLA.