MongoDB has released the source code of mongot, the engine that powers MongoDB Search and Vector Search, under the Server Side Public License. Analysts say the move would help developers of the self-managed version of the database plan better RAG systems for AI use cases, as the code will provide more transparency, debuggability, and control.

By making mongot’s source code publicly available, MongoDB is turning what was previously an Atlas-only (managed version of the database), opaque service into inspectable components, allowing developers to understand how text and vector queries are indexed, executed, and ranked, said Sanjeev Mohan, principal analyst at SanjMo.

The shift is expected to resonate particularly with teams building AI and retrieval-augmented generation (RAG) applications, where visibility into search behavior and failure modes is increasingly critical as systems move from pilots to production, Mohan added.

However, ISG’s executive director of software research, David Menninger, cautioned that developers shouldn’t consider the mongot’s code as open source since it is publicly available.

“Like open-source licenses, the SSPL enables developers to view, use, modify, and share the related source code. It does not meet all the criteria of the Open Source Initiative’s Open Source definition, however, as it requires that anyone incorporating SSPL-licensed code into products offered to an external party (e.g., customer, partner) as a service must release the entirety of their source code for their product under the SSPL,” Menninger said.

But that doesn’t stop developers from using it build applications for their own consumption, said Bradley Shimmin, lead of the data and analytics practice at The Futurum Group.

Rather, the SSPL is “designed specifically” to stop MongoDB’s competitors from taking its free code and selling it as a managed service without paying for it, Shimmin said.

Lowering barriers to adoption

The development could lower barriers to adoption of MongoDB’s offerings, analysts say.

“Previously, if a developer wanted the full MongoDB search experience, they had to be on its managed cloud, Atlas. By releasing the source code, MongoDB is effectively removing the functional wall between their cloud service and their self-managed/Community version,” said Stephanie Walter, practice lead for the AI stack at HyperFRAME Research.

Developers can now test the engines in a local environment, without an internet connection, a credit card, or the need to spin up an Atlas cloud cluster, according to The Futurum Group’s Shimmin.

Analysts say that MongoDB is trying to retain developers with this move, given that the database market is heading towards consolidation, especially around AI applications and use cases.

Typically, most businesses would want to start their AI application development journey on specialized vector databases, but if a developer can test, build, and scale AI systems within MongoDB’s ecosystem, they are less likely to churn, Walter said.

In addition to releasing the source code for mongot, the database provider has also extended the automated embedding capability inside its Vector Search to the Community Edition of the database.

The capability, which automates the process of generating, storing, and updating vector embeddings, reduces complexity for developers when designing a RAG system. Traditionally, developers have needed to construct a pipeline to create and manage vector embeddings, especially for newly ingested data.

Analysts also view the inclusion of the automated embedding capability inside the Community edition as yet another step in MongoDB’s broader effort to challenge rival database providers, especially specialty vector databases.

“This is a direct shot at Pinecone. If the database you already use can handle the complex embedding pipeline for you, there’s really little reason to buy a separate vector-only database,” Walter said.

In addition to Walter, Shimmin believes that the move to add automated embeddings as a capability inside the Community Edition also hurts “glue code” vendors like LangChain.

“It also puts pressure on specialized vector database players to offer more than just storage,” Shimmin added. The automated embeddings capability and mongot are still in preview.