AI Settings — how your agents, skills and models are found
Nothing AI on the mesh is one global list. Each of the three kinds is resolved from a short list of
named sources that belongs to you, stored at {you}/_Memex/AiSettings:
| Kind | Default sources |
|---|---|
| Skills | Platform skills · My skills · Skills of the current space · Skills shipped with the node type |
| Agents | Platform, space and my agents (one exact-membership query) |
| Models | MeshWeaver OpenRouter · Models of the current space · Models shipped with the node type · My models |
Every source is a query template with placeholders, expanded for where you are:
{user}— your home partition{objectPartition}/{objectPath}— the partition / full path of the node you are on{nodeTypePartition}/{nodeTypePath}— the partition / full path of the current node's type
A template whose placeholder has no value where you are is simply not used there (the page says
so), and a query that is not anchored to a partition (namespace: or path:) is never run.
Changing your paths
Open AI Settings: remove a source, add one (name, description, template), or reset a kind to the defaults. The catalogs on the page are scoped by exactly these sources, so what you see there is what the chat resolves for you.
Models
Every model you can see anywhere on the mesh is selectable — selecting adds it to your path as
its own anchored source (the model with its provider); deselecting removes it. You can also
install a model of your own into any space you can write to; it lands at
{space}/Provider/{provider}/{modelId} and that space's models join your path.
The platform's own model path is labeled MeshWeaver OpenRouter. It is served through OpenRouter under MeshWeaver's account — OpenRouter's terms apply — and the model credit included in your plan applies to it. The disclosure lives on the provider node itself and is rendered wherever that path is offered.