Percona has made Percona Search for MongoDB generally available, giving organizations running self-managed MongoDB environments a way to deploy full-text, vector and hybrid search alongside their databases. The offering combines MongoDB’s mongot search engine with automatic embedding generation through local or hosted models, aiming to address a key obstacle for enterprise AI applications: enabling semantic search without requiring sensitive business data to leave controlled infrastructure.
Bringing semantic search closer to enterprise data
The release targets organizations that want to add AI-powered retrieval to applications but cannot easily send internal documents, customer records or other sensitive information to external embedding services. Generating embeddings through hosted APIs can introduce additional data-processing dependencies and usage-based costs, particularly as document collections and query volumes grow.
Percona Search for MongoDB is designed to give database teams more control over that process. Built on MongoDB’s mongot search engine and integrated with Percona Server for MongoDB (PSMDB), it supports full-text, vector and hybrid search within self-managed deployments. The product is now generally available for production use.
The architecture separates database storage from search processing. Percona Server for MongoDB continues to store application data, while mongot maintains search indexes and handles search queries. Applications continue connecting through MongoDB’s normal database interfaces rather than communicating directly with the search process.
This arrangement gives developers access to different retrieval approaches. Full-text search finds documents based on keywords and relevance, while vector search uses embeddings to identify content with similar meaning. Hybrid search combines the two, which can be useful when applications need to retrieve conceptually related information without overlooking exact terms, product names or other specific identifiers.
Local embeddings address data-residency concerns
Automatic embedding generation is a central part of Percona’s announcement. Instead of requiring applications to manage a separate embedding pipeline, mongot can generate vectors when documents are indexed, update embeddings as records change and embed query text during vector searches. Percona Search for MongoDB
The product supports local embedding providers through an OpenAI-compatible interface, including Ollama, vLLM and LocalAI, as well as Hugging Face Text Embeddings Inference (TEI) models deployed within a cluster. This gives organizations the option to run embedding generation on infrastructure they control. Percona says that, with local providers configured, document text can remain within the cluster and search can operate without outbound network access.
That capability is particularly relevant to air-gapped deployments and businesses operating under strict data-residency or internal security policies. It also changes the cost equation: local inference avoids per-token charges from hosted embedding APIs, although organizations still need to account for the compute capacity and operational resources required to run models themselves.
Hosted services remain an option. Percona’s documentation describes support for hosted embedding providers, including OpenAI and Hugging Face Inference API, while the announcement also identifies Azure OpenAI. The choice allows teams to balance infrastructure control, model availability, performance and operating costs rather than committing to one embedding vendor.
Kubernetes operations and production readiness
Deployment and scaling are another focus. Percona Search for MongoDB works with the Percona Operator today, allowing the search tier to be deployed alongside the database and scaled independently. That separation matters because search workloads can have different resource demands from routine database operations.
However, not every part of Percona’s planned operational integration is available at launch. Support for Percona Backup for MongoDB (PBM) and Percona Monitoring and Management (PMM) remains on the roadmap. The planned integrations are intended to bring search-index backup, recovery and monitoring into the same operational workflows database teams already use.
The distinction is important for production deployments. Search indexes are maintained separately from the database’s primary data files, so organizations need a strategy for recovering indexes and monitoring search health. Percona has identified these capabilities as areas it intends to address through its backup and monitoring products, rather than features that should be assumed to be generally available today.
Security is also part of the positioning. Percona says LDAP, OIDC, Kerberos, audit logging and encryption at rest are included in Percona Server for MongoDB without a separate license charge. These are database security capabilities; enterprises should still evaluate how their own authentication, access-control and compliance requirements apply across the complete search deployment.
A different route to AI search infrastructure
The launch reflects a broader infrastructure question facing enterprise AI teams: where should retrieval and embedding workloads run, and who controls the data involved? Retrieval-augmented generation (RAG) applications and AI agents depend on finding relevant information before a model generates an answer or takes an action. Keeping retrieval close to operational data can reduce the need to maintain a separate search system and move information between platforms.
Percona’s approach is aimed at organizations that prefer self-managed infrastructure and want flexibility over embedding models and deployment locations. It also extends the role of MongoDB-based environments beyond transactional data storage into semantic retrieval for knowledge bases, internal applications and AI-powered workflows.
The commercial model is designed to avoid a paid feature tier. Percona says the search product is free to use, with no deployment limits or license-gated features. Its code is source-available under the Server Side Public License (SSPL), rather than an unrestricted open-source license. Customers can also purchase an optional Percona Support contract add-on covering the search tier and related Percona components.
For enterprises, the immediate appeal is control over embedding infrastructure and the ability to introduce search without adopting a separate hosted search service. The longer-term test will be operational maturity, particularly how backup, recovery and monitoring support develop. Those capabilities will help determine whether local-first search can move from an attractive architecture to a dependable foundation for production AI applications.
Percona Search for MongoDB is generally available now. Its documentation lists compatibility with Percona Server for MongoDB 8.3 and later.
Market Landscape
Vector search is becoming a core component of enterprise AI infrastructure, particularly for retrieval-augmented generation, recommendation systems and knowledge discovery. But deploying it in production involves more than selecting an embedding model: organizations must consider data residency, inference costs, index synchronization, recovery, scaling and monitoring.
Percona is positioning its offering around infrastructure control rather than a managed cloud search service. Its local embedding options and Kubernetes-based deployment target teams that want semantic retrieval inside their existing MongoDB environments. The trade-off is that organizations take responsibility for the underlying infrastructure and local model operations.
The planned PBM and PMM integrations will be important to watch. Backup and monitoring are central to operating a separate search tier reliably, and their availability could influence adoption among enterprises with demanding production requirements.
Top Insights
- Percona Search for MongoDB adds full-text, vector and hybrid search to self-managed Percona Server for MongoDB deployments.
- Local embedding providers can keep document text inside a controlled cluster and avoid hosted API token charges.
- The Percona Operator supports deployment and independent scaling of the search tier alongside the database.
- Backup and monitoring integrations with PBM and PMM are planned, not generally available at launch.
- The product is free to use and source-available under SSPL, with optional commercial support.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
