A spreadsheet UI for LLM pipelines and vector RAG

What it does

JamAI Base is an open-source backend that pairs an embedded SQLite database with LanceDB vector storage and adds built-in LLM orchestration, embeddings, and reranking. You interact with it through a spreadsheet-like UI or a REST API, defining tables that generate data, handle real-time app interactions, store knowledge for retrieval, or power chatbots. The idea is to skip procedural backend code and instead declare what you want each column or table to produce.

The interesting bit

The project treats the spreadsheet grid as a declarative programming surface for AI workflows. Rather than scripting ETL pipelines or API handlers, you define table schemas and let the system manage prompt execution, vector search, and context injection—essentially turning cells into serverless functions with memory.

Key highlights

• Four table types cover most LLM app patterns: Generative (auto-fill columns via LLM), Action (frontend-backend bridge), Knowledge (document store with RAG), and Chat (context-aware bots).
• Embedded stack runs locally: SQLite for structured data and LanceDB for vector embeddings, with a serverless architecture that the project claims scales seamlessly.
• Built-in RAG plumbing includes hybrid search, reranking, query rewriting, adaptive chunking, and BGE M3 multilingual embeddings.
• Supports multiple LLM providers—OpenAI GPT-4, Anthropic Claude 3, and Meta Llama 3—through the same declarative interface.
• Self-hostable or available as a managed cloud service with free LLM tokens.

Caveats

• The repo includes a v1-to-v2 migration guide, so early adopters may have weathered breaking changes.
• The contributing guide link is currently a placeholder (Contributing_Guide_Link), suggesting some documentation gaps.
• At 1,098 stars, the project is still building community; long-term maintenance velocity is unclear.

Verdict

Worth a look if you want to prototype RAG-powered apps or internal tools without assembling vector DBs, embedding pipelines, and REST handlers by hand. Skip it if you need a fully mature, battle-tested backend or prefer imperative code over spreadsheet abstractions.