CYROCK.AI DB

The Unified AI Data Engine

Building an AI application usually means running a vector database, a graph database, a metadata store and an agent memory layer side by side - four systems to deploy, connect and keep in sync. CYROCK.AI DB replaces all of them with a single engine, with storage built in. No PostgreSQL, no Redis, no Elasticsearch, no external vector store.

Everything here is driven from one container image: the web console, the REST APIs, the Java and Python SDKs, the Spring AI and LangChain4j connectors, and the MCP endpoint for AI agents.


Capabilities

graph TD subgraph interfaces["How you reach it"] CONSOLE["Web console
:7280"] REST["REST + Swagger
:7281 · :7282"] SDK["Java + Python SDKs
gRPC :7290"] FW["Spring AI · LangChain4j
via the Java SDK"] MCP["MCP for agents
:7285/mcp"] end subgraph engine["One engine"] VEC["Vector search
HNSW · cosine · dot · euclidean"] HYB["Hybrid retrieval
BM25 + vector fusion"] GRAPH["Knowledge graphs
CyQL · traversal · branching"] RAG["Graph RAG
communities · summaries"] MEM["Agentic memory
observe · recall · decay"] META["Metadata
typed, indexed filters"] end STORE["Built-in durable storage
atomic commits · transactions"] CONSOLE --> engine REST --> engine SDK --> engine FW --> SDK MCP --> engine engine --> STORE
  • Vector search and hybrid retrieval - approximate nearest-neighbour search over HNSW indices, fused with BM25 full-text scoring and typed metadata filters.

  • Knowledge graphs - typed, weighted, directed graphs queried with CyQL, a Cypher-inspired language that also covers vector similarity, writes and schema changes.

  • Graph RAG - community detection with hierarchical summaries and global search, plus graph branching so you can experiment on a copy.

  • Agentic memory - an observe/recall/decay lifecycle, context-window construction and entity linking for autonomous agents.

  • Durability - every statement is one atomic commit, with multi-operation transactions and rollback.


Chapters

Start at the top; the first chapter gets you running in a few minutes.

Chapter What it covers

Getting started

Run the container, log in, load sample data, run your first search, query and REST call.

Concepts

Collections and documents, graphs and nodes, vector fields, metadata, projects and roles.

Web console

A tour of the console: sample datasets, the workbench, graph visualization.

CyQL

The query language, by example: matching, traversal, vector similarity, paging, CSV loading.

REST API

The auth model and a curl walkthrough, plus the built-in Swagger UI.

Java SDK

Adding the dependency, connecting, and CRUD, search and transactions.

Python SDK

Installing the package, connecting, and the same API from Python, sync or async.

Framework connectors

Spring AI VectorStore and LangChain4j EmbeddingStore, plus the graph retrievers.

MCP for AI agents

Connecting Claude and other MCP clients, and the tool catalogue.

Configuration

Environment variables: storage, memory, embedding and LLM providers, telemetry.

Operations

Volumes and backup, health and metrics, logs, upgrading, sizing.

Troubleshooting

First-start timing, memory, port conflicts, Apple Silicon, credentials.

Release notes

What is in this release, and its known limitations.


Support and feedback

We want to hear what does and does not work for you. Bug reports, rough edges, missing capabilities and "this was confusing" are all equally welcome.

Both channels live in our GitHub repository:

  • Issues - something is broken, or behaves differently from what this manual says.

  • Discussions - questions, ideas, "how would I model this", and anything you are not sure is a bug.

Both channels are public, so please keep anything confidential to your organization out of a report - and out of the logs you attach to one.