The Agentic Query Layer (AQL).
A software design pattern for giving large language models safe, permission-aware access to structured data. Plain questions in, structured answers out, without the model ever touching the data directly.
Where the layer sits.
The Agentic Query Layer (AQL) sits between your data and a large language model (LLM). Instead of handing the model direct access to a database, you put a controlled layer in front of it. The model asks questions through the layer; the layer queries your systems through their existing APIs and returns structured results. The model never touches the data directly.
Five rules that define the pattern.
01
Plain language in, structured data out
A person asks a question in everyday words. The layer turns it into precise, structured queries and returns clean results with context — no SQL, no report builders.
02
Permission-aware
Access is scoped to what a user is allowed to see. The layer enforces those boundaries on every query, so nobody gets more than they should.
03
Access-controlled
The layer enforces permissions on every request. A connector can support retrieval, changes, or both when the organization's rules allow it.
04
The model never touches the data directly
Data flows one way: out of your systems, through the layer, as results. The underlying data is never exposed to the model, and the model can never modify anything.
05
Dataset-agnostic
AQL is a pattern, not a product tied to one system. Any dataset with an API can plug into it. It just needs a connector.
One pattern. Any dataset.
AQL is not a product tied to one system. It is a pattern — an architectural approach that works on any dataset with an API. Build a connector, plug it in, and the same permission-aware, access-controlled, plain-language doorway works the same way.
Moodle MCP Server — where the pattern starts.
AQL is a pattern, and the pattern only becomes useful when it has a real connector for a real dataset. Our first is Moodle MCP Server — an open-source MCP server that gives the layer permission-aware access to Moodle LMS through its Web Services API. No plugin required on the Moodle side, no direct database access.
Open source core
Install the free core on your own infrastructure and start querying Moodle data right away. It ships with a set of built-in tools:
Agent Edge
The commercial harness built on top of the open core. Turnkey and brandable — call it whatever you like — with an always-growing list of new tools and plugins so it works out of the box, or let us run it for you.
The same layer, built for any dataset.
Moodle is the example because it is everywhere — and because the problem it solves is common. But the pattern is not limited to learning systems. Any organization that holds valuable, structured data behind an API faces the same question: how do we let a model work with this without giving it the keys?
A CRM. A ticketing system. A registry, a ledger, a human resources platform. If the data lives in a system with an API, the Agentic Query Layer (AQL) pattern can sit in front of it — the same permission-aware, access-controlled, plain-language doorway, scoped to a specific dataset.
The result is the same shape no matter what is behind the doorway: staff stop digging through reports and spreadsheets, and start asking the data questions directly — while control stays where it belongs, with the organization that owns the data.
We build the Agentic Query Layer (AQL) for any dataset.
If your data is stuck inside a system your people have to dig through, we can build the connector that gives it a safe way to answer questions directly. Tell us what you are working with.