Today, SurrealDB announced two major milestones: the general availability of SurrealDB 3.0 and a $38 million Series A, bringing total funding to $44 million.

This reinforces a belief we’ve held since we first partnered with Founders Jaime and Tobie: the AI era requires a fundamentally different kind of operational database.

The real bottleneck in AI systems

AI agents are getting better quickly. What’s lagging is durable state.

They forget facts.
They lose relationships.
They struggle to maintain context as workflows expand.

Under most agentic systems today is a stack assembled from separate parts: a relational database for transactions, a vector store for embeddings, a graph database for relationships, search layered on top, and business logic scattered across services. It works, but it’s fragile. Context drifts. Complexity compounds.

If agents are going to move from impressive demos to dependable systems, memory and context have to be treated as core infrastructure.

That’s the problem SurrealDB 3.0 is built to solve.

A database built for agent memory

From the beginning, SurrealDB has taken a multi-model approach, bringing relational, document, graph, time-series, vector, search, and key-value workloads into a single engine with one query layer.

With 3.0, that foundation becomes production-grade.

The core engine has been redesigned with durability and operational predictability in mind. Writes are durably acknowledged by default. Metadata and internal references have been reworked to support safer schema evolution. Index encoding has been standardized to eliminate subtle correctness issues. Client-side transaction support gives application logic a more natural role in transactional guarantees.

There’s also a new in-memory engine with a non-blocking transaction model that increases concurrency and throughput, while enabling time-travel queries for auditing and debugging.

Most importantly, 3.0 brings first-class agent memory and context graphs directly into the database. Models can run close to the data. Structured data, vectors, and relationships coexist natively. Context stays in sync.

Instead of stitching together multiple systems, developers can manage durable agent state in one place.

From storage layer to logic layer

Another meaningful step is the introduction of Surrealism, a programmable control and logic layer embedded in the database itself.

Business rules, access controls, versioning, and AI-driven workflows can live inside the database runtime. Queries can trigger similarity search against vector indexes, enforce assertions at write time, or coordinate AI-powered processes.

When memory and logic sit alongside the data, systems become easier to reason about and more reliable in practice.

The boundary between database and runtime is shifting in the AI era, and SurrealDB is building for that reality.

Why this matters

Every major compute shift reshapes the data layer. AI is no exception.

If agents are going to operate reliably in production, maintaining context across structured and unstructured data while making decisions in real time, the underlying database has to evolve with them. It needs to unify data models, treat vectors as native, support rich relationships, and uphold strong transactional guarantees.

SurrealDB is moving decisively in that direction. The project has built one of the fastest-growing open-source communities in the database ecosystem, with tens of thousands of GitHub stars and millions of downloads. At the same time, adoption in serious production environments continues to grow.

The AI era won’t be defined by models alone. It will be defined by the infrastructure that gives those models durable memory and dependable context.

We’re proud to continue supporting SurrealDB as they build toward that future.