Superduper is an open-source framework for building custom AI applications and agents directly on top of your existing databases. Instead of moving data into a separate AI stack, it brings models to the data: you can run inference, vector search and agent workflows inside databases such as MongoDB and SQL systems, using Python. The Apache-2.0 framework has more than 5,300 stars on GitHub and was actively developed through 2025, and the company behind it now also offers an enterprise agent product.
Most AI projects start with a data plumbing problem โ extracting records into vector stores and keeping copies in sync. Superduper’s answer is to make the database itself the AI platform: models are registered as components and applied to data in place, outputs are stored back alongside the source data, and listeners keep results fresh as records change. That enables retrieval-augmented generation, semantic search and multimodal inference over live business data without a separate pipeline, and the same framework supports building end-to-end applications and agents in Python with frameworks such as PyTorch and Transformers. On top of the open-source core, Superduper the company now markets Superduper Agents, an enterprise product that answers questions and runs analyses across databases, systems and documents for BI and data operations. Together they cover both developers who want an open framework and organisations that want a managed agent layer.
Freemium. The Superduper framework is free and open source under Apache-2.0; the company’s enterprise agent platform is a commercial offering sold to organisations.
Python developers and data teams who want to add LLM features, search and agents to applications without re-platforming their data, and enterprises exploring agent-driven analytics.