Work across several projects without mixing them
Use casesSeveral bodies of work, one machine, one agent. The risk is not losing material — it is answering one client's question out of another's notes.
The situation
A consultant holds material for three clients and their own practice on one machine. Meshes are named groups of vaults — organization and scoping, not an enforced visibility boundary. A search is pod-wide unless you scope it; a search narrowed to one mesh reads the vaults home to it and the vaults it subscribes to, and nothing else. Scoping the question to a client's mesh is how you keep that client's answer inside that client's material.
A mesh is topology, not permission — it decides what a scoped question reaches, not who may open a vault. The distinction is set out in meshes.
How you do it
Start by looking, not changing. Get the groups in front of you and have the agent state exactly which vaults a scoped question would read — and which ones it would leave alone.
Needs: Lyt initialized.
Using Lyt, list every vault registered on this machine with the mesh it belongs to, grouped by mesh. Then tell me exactly which vaults a search scoped to the mesh client-acme would read, and which of my other vaults it would leave untouched. Look only — do not change any vault or mesh.
Read-only. With the groups in front of you, a scoped question is lyt search "rate limits" --mesh client-acme — the vaults home to that mesh, plus the ones it subscribes to. Assumes Lyt is installed and initialized on this machine.
lyt vault list --jsonWhat changes
One machine holds several bodies of work, and each scoped question is answered from exactly one of them.