SurrealDB

What SurrealDB actually is

SurrealDB is a multi-model database sold both as a managed cloud (SurrealDB Cloud, now branded Surreal Cloud) and as a self-hostable engine. One storage engine covers relational tables, JSON documents, graph relationships, time-series, geospatial, and key-value data, queried through SurrealQL — a SQL-shaped language with JSON-ish syntax — and, per the current feature matrix, GraphQL. On top of the core database it bundles full-text and vector search indexing, live query notifications over WebSocket, reactive materialised views, ACID transactions, and fine-grained permissions down to the field level. The pitch in our view is consolidation: one system standing in for a database, a search engine, and a graph store.

The company is SurrealDB Ltd, registered in England and Wales and headquartered in London, founded by Tobie and Jaime Morgan Hitchcock. SurrealDB was released publicly in 2022, hit 1.0 on 13 September 2023, and shipped version 3.0 in February 2026. It has raised $44M to date — a $6M seed in January 2023 and a $38M Series A in November 2025 — which matters when you are betting a production workload on a young database: the company is funded well past the seed stage but is still small, with a team of 39 listed on its about page. Reference customers named publicly include Samsung, Nvidia, Verizon, Tencent, and Walmart.

Licensing is worth reading before you commit. The database engine is free to self-host under the Business Source License 1.1 (BSL), with SDKs and tooling under Apache 2.0 or MIT. BSL is not an open-source license in the OSI sense — it restricts offering the database as a managed commercial service — but for running your own workloads it is permissive enough in practice. If a true open-source engine is a hard requirement for your organisation, this is the first thing to check with your legal team.

Pricing: what you really pay

SurrealDB's pricing page publishes a complete rate table for its entry Cloud plan, which we appreciate — most database vendors hide the top end. The Start plan is pay-as-you-go, billed by the hour: the first instance is $0/hr and includes 1 GB of storage free forever, with 1 GB per month of internet egress free. Further instances start at $0.021/hr, which the same page computes as $15.33 per month at 730 hours for a Small (0.5 vCPU, 1 GB memory). The ladder continues: Medium (1 vCPU, 2 GB) at $0.044/hr or $32.12 per month, Large (2 vCPU, 4 GB) at $0.093/hr or $67.89 per month, and general-purpose tiers up to a 4X Large (16 vCPU, 64 GB) at $2.409/hr, about $1,758.57 per month. Storage beyond the free gigabyte runs $0.239 per GB up to 512 GB, and compute and storage are billed separately.

The Scale plan is for production workloads that need high availability: billed by the node-hour at $0.192 per node, with a three-node minimum, promising multiple-availability-zone deployment, horizontal scaling, and multi-region disaster recovery. Note the multiplier — the minimum buy-in is three nodes, so the real floor is roughly three times the hourly node rate before storage. Enterprise is priced per deployment for self-hosted or dedicated single-tenant installs, with custom SLAs, FIPS-compliant cryptography, audit logging, and support tiers; annual plans across Cloud and support are discounted but quoted by sales.

Our read: the free instance is genuinely usable for prototypes and small services — 0.25 vCPU and 1 GB memory is a real machine, not a token sandbox — and the hourly billing means a development instance costs pocket change if you scale it down. The trap, if there is one, is the jump to Scale: pricing is configuration-dependent and node-count-driven, and the feature list marks several headline items (database branching and forking, object-storage-backed data) as "coming soon" on both Start and Scale. Do not buy a plan today for features marked coming soon.

Performance & infrastructure

SurrealDB Cloud runs on AWS, and the surreal cloud catalog regions command in the official docs lists exactly three deployable regions: aws-euw1 (Europe, Ireland), aws-use1 (US East, N. Virginia), and aws-usw2 (US West, Oregon). That is a thin footprint — there is no APAC region, and AWS PrivateLink private connectivity is currently available only in US West (Oregon). Instances can be created public, private, or dual access mode, and the CLI supports estimating cost before you create anything, restoring from snapshots, and pausing instances — details we like, because they show the operational surface is thought through.

On the engine side, SurrealDB's pitch is one query layer over many data models, and the feature matrix backs it up: ACID transactions, schemaful and schemaless operation, full-text and vector search indexing, time-series and geospatial storage, and live queries that push updates to clients over WebSocket — the real-time subscription story is native rather than bolted on. The permission model is unusually granular for a young database: access rules can be defined at the table, document, and field level directly in the schema, with RBAC and ABAC in the Cloud control plane.

We have not benchmarked SurrealDB ourselves, so we take no position on its throughput relative to PostgreSQL or MongoDB. Third-party commentary consistently describes performance as improving fast but the ecosystem as early; treat any vendor or community benchmark you find as marketing until it matches your own workload shape, and compare against your current stack using our benchmarks methodology before migrating anything that matters.

Where it falls short

The biggest honest caveat is maturity. Version 1.0 shipped in September 2023, and 3.0 only arrived in February 2026. Postgres has decades of operational scar tissue; SurrealDB's storage engine story is still actively changing — the company recently announced it is replacing RocksDB with its own SurrealKV engine and introducing an in-memory SurrealMX engine. That is impressive engineering velocity, and it is also churn in the layer you least want churn in. In our view, running a mission-critical primary database on SurrealDB today means accepting that storage-engine migrations and breaking changes are part of the ride.

Ecosystem is the second gap. The tooling and integration surface around PostgreSQL or MongoDB — BI connectors, ORM coverage, DBA tooling, hiring pool — is far larger. SurrealDB's own feature matrix concedes the point in places: object-storage-backed data is listed as coming soon, as are database branching and forking outside Enterprise. Analytics and BI integrations are thin, which matches what we hear from teams who tried pairing it with existing reporting stacks.

Third, support and SLA honesty. The publicly documented plans give you community support on the free tier and progressively paid support above it; custom SLAs are an Enterprise conversation. For a managed database, we would want a published availability number on the paid Cloud plans. There is also a small documentation wrinkle worth knowing: the Start plan lists daily automated backups, but third-party write-ups of the product FAQ claim automated backups with retention apply to paid tiers — if backups are load-bearing for you, confirm the retention policy in writing before deploying.

Who should pick this

Our pick for teams building real-time, multi-model applications who would otherwise be gluing a document database, a search engine, and a graph store together: collaborative apps, personalisation and recommendation backends, knowledge-graph style products, and anything where live queries and row-level permissions are first-class needs. The free instance plus 1 GB of storage is enough to build a working prototype, and hourly billing keeps idle development environments cheap. Teams with GDPR constraints can keep data in the Irish region, and the self-host option under BSL 1.1 is a genuine escape hatch if the cloud pricing stops making sense.

We would not pick it as a drop-in Postgres replacement for a conservative production workload, nor for teams that need APAC regions today, nor for organisations whose compliance regime requires a true open-source database engine. If you are evaluating it against MongoDB Atlas or PostgreSQL, the deciding question in our view is not raw query performance — it is whether consolidating your data layer saves you more operational complexity than the immaturity of the platform costs you.

Key takeaways

• Multi-model engine (relational, document, graph, time-series, geospatial, key-value) with built-in full-text and vector search and native live queries.
• Free Cloud instance (0.25 vCPU, 1 GB memory, 1 GB storage) with hourly billing from $0.021/hr; published rate table up to $2.409/hr for 16 vCPU / 64 GB.
• Scale plan is node-hour billing at $0.192 per node with a three-node minimum — budget accordingly for high availability.
• Only three AWS regions today (Ireland, N. Virginia, Oregon); no APAC region and PrivateLink limited to Oregon.
• Young product: 1.0 in September 2023, 3.0 in February 2026, and the storage engine is still being replaced — plan for churn.
• Engine is BSL 1.1 (source-available, not OSI open source), self-hostable, with SOC 2 Type 2, ISO 27001, and Cyber Essentials Plus certifications on the company.

Sources

• https://surrealdb.com/pricing
• https://surrealdb.com/about
• https://surrealdb.com/releases/1.0
• https://surrealdb.com/docs/reference/cli/cloud/commands/catalog
• https://surrealdb.com/docs/explore/cli/cloud/instances
• https://surrealdb.com/docs/manage/instances/private-connectivity