Get started

Install the memQL Cockpit, connect to an engine, and run your first agent. This takes about five minutes.

This page is written for the site and is pending an official version in the memQL engine repository. The install command is real and hosted at memql.io/install.sh; it requires a published memQL Cockpit release to download.

#Requirements

memQL Cockpit currently supports macOS and Linux. There is no Windows installer yet; on Windows, use WSL2 (a Linux environment) in the meantime.

• macOS 13+ (Apple Silicon or Intel), or a modern Linux distribution (x86-64 or arm64)
• A terminal

#Install

Run this in your terminal — it downloads and installs the Cockpit:

curl -fsSL https://memql.io/install.sh | sh

This drops the memql command on your PATH. Verify it:

memql --version

#Your first agent

1. Start the Cockpit.

   shell⧉copy

   memql cockpit

   The Cockpit is the terminal-native console for driving memQL — clusters, chat, concepts, workers, and safety in one place.

2. Connect to an engine. On first launch the Cockpit walks you through pointing at an engine (local or remote) and authenticating.

3. Write your first memory. From the chat tab, send a message. memQL records it as an episodic memory in the graph — durable across turns and restarts.

4. Run a turn. Ask the agent to do something. Watch the agent loop run in the Cockpit: plan, step, observe — with the cost and safety spine bounding it.

5. Recall. Ask a follow-up. The agent recalls what it learned rather than starting from scratch — that is the memory substrate doing its job.

#Where to go next

• Why it's a harness, not a library — the proof behind the positioning.
• The language (MQL) — every behavior in the system is a typed construct.
• Concepts — the data model, events, and the graph the memory lives in.