Feature Proposal: Integrate OceanBase seekdb as AI-Native Database

[Teingi]

Feature Proposal: Integrate OceanBase seekdb as AI-Native Database

Summary

I propose integrating OceanBase seekdb as the primary database for LemonAI to enhance knowledge retrieval, semantic search, and AI-native capabilities.

Motivation

Currently, LemonAI uses SQLite for data storage and implements basic text-based search for knowledge retrieval. While this works for basic use cases, the project could significantly benefit from an AI-native database that provides:

1. Semantic Search: Currently, knowledge retrieval uses simple text matching (includes()). seekdb enables semantic search to find relevant knowledge even when exact keywords don't match.

2. Vector Search: The codebase has commented-out vector functionality (syncQuestionVectorData), indicating interest in vector capabilities. seekdb provides built-in vector search.

3. Hybrid Search: Combine vector search, full-text search, and relational queries in a single statement for more accurate results.

4. In-Database AI: Run embedding generation, reranking, and LLM inference inside the database, reducing external API calls.

5. Better Knowledge Retrieval: The current resolveUsedKnowledgeForCategory only filters by category and agent_id. With seekdb, we can find semantically similar knowledge across categories.

Why seekdb?

Key Advantages

Feature	Current (SQLite)	seekdb
Vector Search	❌	✅
Semantic Search	❌	✅
Hybrid Search	❌	✅
Full-Text Search	⚠️ Limited	✅
Embedded Mode	✅	✅
MySQL Compatible	❌	✅
In-Database AI	❌	✅
OLTP Support	✅	✅

Perfect Fit for LemonAI

1. Embedded Mode: seekdb supports embedded mode (like SQLite), perfect for local deployment
2. MySQL Compatible: Easy migration path, minimal code changes
3. Unified Engine: Handles relational, vector, text, JSON in one database
4. RAG Workflow: Built-in support for document-in/data-out RAG workflows
5. Apache 2.0 License: Open source and compatible

[Teingi]

References

• seekdb GitHub
• seekdb Documentation
• Python SDK