znasllc-io/memql

**Build applications where AI can work with your data, tools, and workflows.**
MemQL is an open-source AI platform. Its engine combines typed, versioned
records and relationships with queries, mutations, model routing, agent work,
and event-driven automations. A single .memql language connects those pieces.
Integrations bring in external systems; hosted sites and SDKs give people a way
to use what you build.
> **Alpha / pre-1.0 — not production-ready.** Expect breaking changes to the
> language, engine API, and wire protocol. Start with experiments and prototypes.
Get started ·
Documentation ·
Visual Studio Code and Cursor extension ·
MemQL OS
One engine, several ways to work
| Part | What it does | Where you use it |
|---|---|---|
| **MemQL engine** | Stores the memory graph, enforces declared access rules, executes constructs, routes model calls, and records agent work | A cluster, reached through gRPC or the browser WebSocket bridge |
| **MemQL OS** | The browser workspace for managing that cluster and working with its data | Apps such as Fleet, Files, Deployables, Nexus, Concepts, and Logs |
| **MemQL for Visual Studio Code and Cursor** | Edits .memql files offline; connects to clusters to inspect, run, and train constructs | Your editor |
| **Your applications** | Present your own product using MemQL's capabilities | A hosted site or an external client built with an SDK |
The engine is built on a time-series memory graph backed by PostgreSQL,
TimescaleDB, and pgvector. That storage supports the platform; MemQL also runs
the behavior around it. The agent harness is its durable work system, one part
of the broader product. Explore the capabilities.
Start with something you can inspect
A concept describes a kind of record. A query names a reusable read operation.
Here is a caller-owned reading list:
@actor supplies the authenticated caller. The query filters to that caller,
and @rowAuthz declares the ownership tier. paginate 50 bounds the first
page. The complete reading-list tutorial
adds a mutation and a tool, explains how to validate the file, and walks through
running it in VS Code or Cursor. Saving a file does not deploy it.
Get started
- **Try the language:** build and install the Visual Studio Code and Cursor extension, then open the example. Editing works without a cluster.
- **Use an existing cluster:** add it in the extension, sign in, and inspect a query before running it.
- **Run MemQL locally:** follow the quickstart. The supported local stack uses Docker, k3d, and ArgoCD. An initial image build can take time.
A visual workspace for the cluster
MemQL OS is a single-page application made of focused apps. Fleet manages
machines and execution resources; Files holds artifacts; Deployables manages
what is served; Nexus exposes goals, runs, and approvals. It uses the same
engine APIs as other clients.
**Supervised Visual Composition**
is the approved design direction: compose objects and their relationships
with mouse and keyboard, and review MemQL's proposals in the same visible
workspace. Fleet provides visual composition, persistent routing-policy editing,