HydraDB vs SurrealDB
SurrealDB is one database for every data shape: relational, document, graph and vector in one engine and one transaction. Your team models the graph, or sends conversations and documents to its Agent Memory. When agents need a graph built from Slack, email and docs that keeps evolving, with their memory and the company's knowledge in the same database, HydraDB is worth a look.
Based on public docs and pricing as of Nov 2026. Something out of date? Suggest a correction →
SurrealDB is the pick for teams that want relational, document, graph, vector, time-series and full-text data in one engine, queried together in one ACID transaction.
It's written in Rust, runs as a server or embedded in an app or browser, and SurrealDB 4 speaks the PostgreSQL and MySQL protocols. It's the go-to for app teams who'd rather not run Postgres, a vector database and a graph database side by side.
The friction shows up when nobody has modelled the graph yet, and the knowledge an agent needs is spread across apps rather than sitting in your tables.
The caveats worth knowing
Your schema is your ontology. You define the tables and create each edge with RELATE. Turning a Slack thread or a design doc into people, projects and decisions is extraction code you write and run.
Apps don't come in on their own. SurrealDB stores what you send it. There are no connectors for Slack, Gmail, Drive or Jira, so syncing them, and their permissions, is your pipeline.
History depends on how you start the database. VERSION queries return a record as it was at a past time, on the SurrealKV engine with versioning enabled at startup. Why a fact changed, and where the new value came from, are yours to store.
What changes with HydraDB?
The graph is created from your data, and keeps evolving. Docs, chats and tickets are turned into entities and relationships, structured to your ontology or one proposed for you. As new data arrives, the graph updates incrementally and restructures when new kinds of facts appear.
Sources connect directly. 100+ connectors keep Slack, Gmail, Drive, Jira and others in sync, permissions included.
Agents can find what changed, when, and why. Every change to the graph is versioned like code in Git, with when, why and where it came from.
What about SurrealDB Agent Memory?
Agent Memory is SurrealDB's memory product, built on the database and launched in October 2026. Send it a conversation turn, a document or a fact, and a language model turns it into typed entities, attributes and relations. Agents use seven MCP tools, including remember, recall, reflect and forget, or the REST API and Python and TypeScript SDKs.
It's the go-to for agents that need to remember their own conversations and the documents users hand them, on a database the team already trusts.
The difference shows up when an agent needs what it remembers and what the company knows in the same answer.
Memory lives in the same database as knowledge. Agent Memory keeps each memory context in its own namespace and database, apart from your application data. In HydraDB, an agent's memories and the company's knowledge live in the same database and come back together in one query.
The company's apps come in too. Agent Memory takes conversation turns, facts and uploads, with no connectors to workplace apps. HydraDB syncs 100+ apps into the same knowledge graph.
The billing units differ. Agent Memory counts tokens on every model and embedding call, background work included: $300 a month buys 10M tokens on Standard, about $30 per million. HydraDB charges $0.70 per 1M tokens ingested.
Technical comparison
| SurrealDB | SurrealDB Agent Memory | HydraDB | |
|---|---|---|---|
| What it is | Multi-model database | Memory service built on SurrealDB | Graph infrastructure: storage, creation, evolution and retrieval |
| Storage | Written in Rust; SurrealKV on disk, in-memory, or distributed storage on one object-store bucket | Closed-source layer on SurrealDB | Written in Rust; the graph lives on object storage, with memory and NVMe as caches |
| Data models | Relational, document, graph, vector, time-series, full-text and geospatial | Entities, attributes and relations extracted from what you send | Property graph with vector and BM25 indexes; memories and knowledge in the same database |
| Graph creation | You define tables and RELATE each edge | A language model extracts from conversations and documents; same entity under different names isn't merged automatically | From 100+ app connectors, file uploads and the API, structured to your ontology or a proposed one |
| Keeping it current | Your writes; schema changes are yours | New information is extracted and reconciled as it arrives | Updates incrementally as data arrives and restructures when new kinds of facts appear |
| Retrieval | SurrealQL: graph traversal, HNSW and DiskANN vector indexes, BM25, rank fusion | Recall and context tools over the extracted memory | Vector search, BM25 keyword search and graph traversal combined in one query, with metadata filters and recency weighting |
| Query interface | SurrealQL; PostgreSQL and MySQL protocols in 4.x | MCP, REST API, Python and TypeScript SDKs, CLI | Natural language through the API and SDKs; openCypher on the open-source database |
| When a fact changes | VERSION queries on SurrealKV with versioning enabled | Not documented | Every version kept, Git-style, with its source |
| Deployment | Embedded, self-hosted, or Surreal Cloud on AWS or Azure | Managed; single-tenant and air-gapped on Enterprise | Managed, dedicated, your cloud, or self-hosted |
| Pricing | Free tier; Earth Pro $15/month; Cloud instances from $0.021 per node hour (pricing) | Sandbox free (3M tokens once); Lite $30/month (1M tokens) to Plus $1,100/month (40M) | Managed from $25/month: $0.70 per 1M tokens ingested, $0.50 per GB-month storage; open source free (pricing) |
| License | BSL 1.1: no offering it as a database service; becomes Apache-2.0 on 1 January 2030 | Closed source | Elastic License 2.0 |