MariaDB has announced Enterprise Platform 2026 with a strong message for developers, data teams, and DBAs: unify transactional, analytics, and vector engines into a single platform, add native RAG and AI agents, and bring everything to a serverless cloud environment to handle elastic and unpredictable workloads typical of agentic applications. The promise is clear: fewer components, less latency, and more speed when converting operational data into business value.
The company positions this launch as the “ultimate platform” for building the next wave of intelligent applications. Beyond the slogan, the package includes architectural decisions aimed at fundamentally cutting down the classic dance of ETLs, data lakes, separate vector stores, and multiple pipelines that slow teams down.
What does “agentic” mean in practice (and why does it matter to a database?)
The term AI agentic describes agents capable of reasoning, acting, and coordinating tools with high-level objectives. In data, this involves two needs:
- Querying and understanding information with semantic context (hence vector search and RAG).
- Transacting in real-time (creating orders, adjusting prices, opening claims, logging events) without jumping between platforms.
MariaDB 2026 tries to solve both in the same execution plane: transactional + analytical + vectors + RAG, with copilots that expose capabilities via natural language. The idea: if the agent lives “inside” the database and it already understands semantics, statistics, and operations, the jump from a business event to an intelligent, actionable response is shorter.
The new features, piece by piece
1) Native vectors (without “another” vector database)
After introducing vector search earlier this year, MariaDB emphasizes a native approach: no need for another engine for embeddings, which reduces latency and data movement. This proposal saves infrastructure and, more importantly, operational complexity: a single authentication, observability, and security for everything.
2) “RAG-in-a-Box”: managed, automatic RAG
The platform includes MariaDB AI RAG, its “RAG in a box”. The company claims it eliminates the need for retrieval pipelines, vector stores, or even managing embeddings explicitly: the database handles all steps automatically and optimally. For those who have suffered through the “threads” of a classic RAG (tokenization, embedding batches, chunking, refresh), the promise of operationalizing RAG with fewer pieces can be a game-changer.
3) Embedded copilots: a Text-to-SQL for developers and a “DBA as a service”
MariaDB Cloud offers ready-to-use agents:
- A developer copilot (Text-to-SQL) that responds in natural language and generates queries directly on the data.
- A DBA copilot, focused on tuning, debugging, and operational tasks, with an emphasis on team productivity.
The key isn’t the buzzword, but integration with security, catalog, and auditing features of the platform itself.
4) MCP Servers: glue with the agent ecosystem
The integration with Model Context Protocol (MCP) Servers enables agents to communicate with MariaDB and other systems in a standardized way. Besides querying or vector search, these servers can launch serverless databases in the cloud and connect to MariaDB copilots. Result: intelligent automation that reads, reason, and also executes actions (create a database, run a migration, trigger a task).
5) Serverless in MariaDB Cloud: elastic by default
Agentic platforms experience unpredictable peaks. To absorb these without maintaining underutilized machines, MariaDB offers its serverless database: elastic scaling, operational simplicity, and pay-as-you-go. Nothing new in the cloud world, but crucial when 100 agents decide to execute simultaneously or when a RAG flow spikes due to a marketing campaign.
6) Accelerated operational analytics with MariaDB Exa
The most powerful analytics feature is MariaDB Exa, designed for complex analytics on operational data at multi-terabyte scale, and — according to the company — at speeds >1,000× higher than traditional OLTP engines and several times above leading analytical engines. It relies on a strategic alliance with Exasol. The idea: immediate insights without moving data out of the system — in-place queries to minimize friction.
7) Enterprise-grade performance, security, and management
- Performance: in internal benchmarks, MariaDB Enterprise Server 11.8 — the core of the platform — achieves +250% compared to version 10.6.
- Management/observability: Enterprise Manager centralizes topologies, metrics, and offers a visual IDE for queries and schemas.
- Security: MaxScale includes an improved database firewall with programmable rules to control how each user queries and to reduce attack surfaces.
Why combine OLTP, OLAP, and vectors in the same “box”?
Many teams have discovered that the time lost moving between transaction logs, ETLs, data lakes, vector stores, and RAG services is where the product fails: latencies breaking UX, rising costs, and a lot of orchestration. MariaDB’s approach is to collapse this path: if data is created in the same platform where it’s analyzed, vectorized, retrieved for a LLM, and acted upon (transacted), then it’s simpler and potentially faster.
Is “single database” always better? Not necessarily. For hyper-scale analytical workloads or organizations with established standards (e.g., specialized data warehouses), a hub-and-spoke approach still makes sense. But for applications where proximity between transactional data, semantics, and action creates value, unification reduces friction and risk of drift between sources.
What benefits (and risks) for developers and DBAs
For developers, the promise is tangible:
- Fewer SDKs and queues: one endpoint, with Text-to-SQL for exploring data and RAG without “sewing” pipelines.
- Lower latency from data to answers: agents retrieve context and act on the same plane.
- More focus on logic and end-user experience.
For DBAs, two main messages:
- Copilot for routine tasks (tuning, debugging, query analysis) with a Enterprise Manager that centralizes observability and operations.
- More surface area to govern: OLTP + OLAP + vectors + agents in the same house demands security discipline, granular access controls, data governance, and resilience testing (partial failures, isolations, quotas).
Risks to watch:
- Governance: with everything together, segmentation (schemas, roles, workspaces) and default security are critical.
- Unpredictable costs in serverless if limits and alerts aren’t set (agents don’t sleep…).
- Coupling: the more we rely on native, proprietary capabilities, the more vital it is to plan portability strategies.
Which use cases fit best?
- Business applications with embedded AI: CRM/ERP with assistants that read, reason, and execute (e.g., create order, adjust credit, open a task).
- Customer support and virtual agents: RAG with real-time operational context, no synchronization “forks”.
- Operational analytics and immediate insights: dashboards and decisions with Exa right on live data.
- Agentic automation: agents launching serverless bases, moving data within pre-defined policies, and documenting actions.
Availability and roadmap
MariaDB Enterprise Platform 2026 is available immediately for clients. The improvements announced for MariaDB Cloud will start rolling out from November 1, 2025. The company encourages trying out the platform in the cloud and participating in webinars about the new features.
What should interested organizations do today?
- Controlled pilot: pick a business case where RAG + transaction bring value (e.g., an internal assistant with limited autonomy).
- Security model: define roles, limits, and traceability for copilot and agent actions (who can do what, how far).
- Observability & costs: enable detailed metrics (latency, caches, vector recall) and spending guardrails in serverless.
- Portability plan: encapsulate agent logic (MCP, SDKs) to minimize coupling and keep options open.
A step closer to usable AI in production
The key aspect of MariaDB 2026 isn’t a single component, but the composition: vectors, RAG, copilots, MCP, and serverless with OLTP/OLAP combined. Is it a silver bullet? Not really. But it’s a shortcut for teams moving from demos to operational services without juggling five different products. Time will tell how much of the promise — less parts, more results — holds at scale. For now, the direction is clear: bring intelligence closer to the data and action.
Frequently Asked Questions
What is “RAG-in-a-Box” and how does it differ from traditional RAG?
It’s the native implementation of RAG within MariaDB: the platform automates steps like segmentation, vector indexing, retrieval, and combination with LLM without teams needing to create and orchestrate pipelines or operate external vector bases. The goal is to reduce latency and complexity.
What are the embedded copilots in MariaDB for?
There are two main ones: a developer copilot (Text-to-SQL) that responds in natural language with queries and insights, and a DBA copilot that assists with performance, diagnostics, and operational tasks. Both reside in MariaDB Cloud and adhere to platform controls and auditing.
What role does the Model Context Protocol (MCP) Servers play?
They serve as a standard bridge enabling agents to interact with MariaDB and other systems, executing advanced operations: beyond queries and vector searches, they can launch serverless databases and coordinate with MariaDB’s copilots, enabling automated workflows.
What performance and analytics does MariaDB Exa provide?
Exa targets complex analytics on operational data at multi-TB scale, delivering speeds >1,000× faster than traditional OLTP engines and multiple times faster than leading analytical engines. It leverages a strategic partnership with Exasol, aiming for immediate insights without moving data out of the system — in-place queries to minimize friction.
Sources: official announcement of MariaDB Enterprise Platform 2026 (unified OLTP/OLAP/vectors layers, RAG-in-a-Box, copilots, MCP Servers, serverless in MariaDB Cloud, MariaDB Exa, partnership with Exasol, security and management improvements; performance of MariaDB Enterprise Server 11.8).**
via: mariadb