AI-native infrastructure · Europe

We write the layer underneath.

A database engine that generates embeddings inside itself. Client SDKs, an MCP server, and a memory layer for AI agents. Written from scratch in Rust — with the measurements published, including the unflattering ones.

Building in the open · numbers you can reproduce
How we work

Engineering, not announcements.

Infrastructure earns trust the slow way. We publish what a thing does, what it costs, and where it breaks — then let the numbers argue.

🧩

One process, not a stack

SQL, vector search and embedding generation in a single engine — no network hop between parts that need each other.

📐

Measured, not claimed

Every number ships with the command that reproduces it — and with the cases where we are slow.

🦀

Written from scratch

Rust, no borrowed engine underneath. We own the parts our products depend on.

What we build

Infrastructure, not wrappers.

A database engine, a memory layer for AI agents, a quantum simulator. Written from scratch in Rust. Each ships with the numbers behind it — including the ones that are not flattering.

🧊Beta

ZCube

The engine. Columnar SQL, vector search and embedding generation in one process — no network hop between the parts that need each other.

  • Columnar SQL over Arrow and Parquet
  • FLOAT[n] vector columns with their own HNSW index
  • Embeddings generated inside the engine
  • Time-travel queries for vector and graph zones
  • Realtime over SSE, WebSocket and WebTransport
  • Ledger, encryption at rest, row-level security
  • A terminal client: SQL/ZQL REPL, migrations tracked by checksum and rolled back, row-level-security policies and seeds scaffolded from the command line
Source on GitHub →
🔌Beta

ZCube SDK

Client libraries. Permissively licensed on purpose — a client ends up inside your code, so it must not carry the engine's terms.

  • TypeScript, Go, Python and Rust clients
  • Adapters for Next, Nuxt, Svelte, Solid, Remix, React Native
  • 33 typed RPCs over gRPC
  • Apache-2.0
Source on GitHub →
🔗Beta

ZCube MCP

ZCube handed to AI agent-builders as a general-purpose database, over the Model Context Protocol.

  • Model Context Protocol, revision 2025-06-18
  • stdio and HTTP/SSE transports
  • The database itself, not a wrapper around one
Source on GitHub →
🧠Early access

ZMemory

A memory layer for AI assistants, built on ZCube. What the assistant learned yesterday is still there tomorrow.

  • 77 tools over the Model Context Protocol
  • Hypotheses you can raise, then resolve later
  • Time-travel across your own history
  • Multi-agent consensus
  • One-command install for five IDEs
Founders access soon