Surrealdb Adds $23M for Agent Memory Database

SurrealDB raised a $23M Series A extension and launched v3.0, bringing total funding to $44M. "Every single thing is transactional," CEO and co‑founder Tobie Morgan Hitchcock said. Read the coverage on VentureBeat.

What it is:

SurrealDB is an open-source, multi-model database that stores documents, graphs, vectors, full-text, time-series, and relational data in one engine with a single query language called SurrealQL. Translation: one database for context, memory, search, and retrieval.

It is ACID-transactional (Atomicity, Consistency, Isolation, Durability), written in Rust, and you can run it yourself or use SurrealDB Cloud. Learn more at the SurrealDB site.

Who’s behind it:

The company is run by brothers Tobie Morgan Hitchcock (CEO) and Jaime Morgan Hitchcock (COO). SurrealDB was founded in 2021 in London. Earlier rounds include a $6M seed from FirstMark and a $20M Series A led by FirstMark and Georgian, with Crew Capital and Alumni Ventures. Details on investors for the new $23M extension were not disclosed. See company info on Craft.co.

Why agents care:

By "agents" we mean AI agents - automated systems that act on your behalf and need memory of past interactions. SurrealDB doesn’t bolt memory on as a separate cache. It persists agent memory as graph relationships and semantic metadata alongside your structured data and embeddings. That means you can query vectors, graph edges, and rows together in a single transactional operation.

In v3.0, synchronized writes are the default, and HNSW vector indexing is sped up with hash-based deduplication. (HNSW is a fast approximate nearest-neighbor index used for vector search.) More on that in the VentureBeat piece.

How it’s different:

Unlike pure vector stores such as Pinecone, Weaviate, or Milvus - or a Redis-plus-embeddings workaround - SurrealDB is a multi-model, ACID database with native vector, graph, and relational features. That reduces the need to juggle multiple systems when building retrieval-augmented generation (RAG) systems and agent memory.

Receipts:

• The v3.0 benchmarks reported big gains: for example, 8-22x faster graph queries and up to 8x faster HNSW vector search.
• Cited users include Tencent and Later.com running production workloads.
• GitHub stars are around 31k.

As always, run your own proof-of-concept tests. Benchmarks and details are available on SurrealDB’s blog.

What you can build:

• Support agent that remembers past tickets and user preferences without re-sending long prompts each time.
• Sales copilot that links CRM records, knowledge-base documents, and call summaries, then recalls context across weeks of interactions.

Stack notes for founders:

• Cost control: store embeddings and provenance once, retrieve selectively, and stop resending giant prompts for every request.
• Integration: self-host or use the managed Cloud; the API is the same either way.
• Competitive angle: fewer moving parts, fewer synchronization bugs, and faster release cycles for product teams.

More context on the company’s goals and cloud offering is in SurrealDB’s fundraising blog post: SurrealDB blog.

Past context for the money:

The 2024 Series A was led by FirstMark (a New York VC) and Georgian (a Toronto growth fund). This extension aligns with SurrealDB’s push to own the “memory layer” as AI agents move beyond demos into production. The company has not shared specific use-of-funds for the extension.