docsOverview

What is MemQL?

MemQL is an open-source AI platform for applications that need AI to act on structured data and external systems. You declare data and behavior in .memql files; the engine executes them on a cluster. You can inspect that work through MemQL OS, Visual Studio Code, Cursor, or your own client.

#The mental model

The engine runs the system. MemQL OS is an application connected to it. Individual OS apps are views and controls for particular jobs, not separate engines. The Visual Studio Code and Cursor extension is another client, with offline language support and an optional connection to one of your clusters.

text

MemQL OS apps     VS Code / Cursor     Your sites and applications
       \                 |                     /
               Authenticated engine API
                          |
  Data and relationships · Tools and automations · Agent work
                          |
        Versioned memory graph and execution records

#What the engine supports

Capability	What you can do	Start here
Typed data and history	Declare concepts, validate fields, query rows, and retain versions and relationships	First program, versioning
Reusable behavior	Declare queries, mutations, logic, tools, prompts, and event/schedule automations in the same language	Language, specifications
AI routing	Route a call's required level through policies and rules to compatible configured sources, including local models	Routing, local models
Durable agent work	Inspect goals, runs, steps, approvals, budgets, and recovery records	Harness, measured results
Files and knowledge	Store versioned artifacts, train from Library files, and retrieve knowledge	Library, document history
External systems	Connect integrations; distinguish mirrored, original, and native data; deliver changes through an outbox	Data origins, outbound delivery
Sites and applications	Serve static bundles by hostname; manage sources, builds, and deployments	Hosting, Deployables
Communication	Use the implemented audio stream and campaign sending surfaces when their providers and delivery configuration are ready	Audio, campaigns
Access and operations	Authenticate users, grant app access, inspect logs, and manage the cluster	Access model, MemQL OS

A capability being implemented does not mean a new cluster has it configured. Model calls need a compatible inference source; mail delivery needs a sender; external integrations need their own configuration and permissions. Use configuration readiness to see what is missing.

#The memory graph

MemQL is built on a time-series memory graph, backed by PostgreSQL, TimescaleDB, and pgvector. Concepts describe record types. Queries select data; mutations write versions. Relationships connect records. Authorization depends on the declared concept tier and the operation: do not assume every concept is private by default. Learn the access model.

The agent harness uses that same system for its work records. It is built into the platform's work lifecycle, not a pack you can switch off. Its evidence is scoped to the tests and runs in the proving scorecard; a replay result is not a guarantee about every external side effect.

#Modules, clients, and apps

Modules describe what an operator can inspect in the engine: components, integrations, packs, and node types. Only packs have the pack enable/disable switch, and changing it takes effect at node restart. Modules.
Clients connect to the engine. MemQL OS is the browser client included in this repository; product-specific clients live in their own repositories. Clients.
OS apps organize a person's work inside that client. Fleet, Files, Deployables, Nexus, and Concepts share a session and cluster connection. Find the right app.

The approved OS design direction is Supervised Visual Composition. That page distinguishes available Fleet composition and policy review from designs still being implemented or awaiting approval.

#What to expect today

MemQL is alpha and pre-1.0. APIs and language syntax can change between releases. The local runtime is a Docker-backed k3d cluster managed through ArgoCD; a cloud installation follows the same GitOps topology. It is a multi-service system, not a single editor process. Offline editing is available before installing it.

Get started, or choose a route from the documentation home. See status and direction for the boundary between implemented capabilities and upcoming work.