Settings
Configure inference optimization and account boundaries.
Settings help each connected app use models deliberately. Keep the default path for most everyday use, then add closer limits when an app or agent needs its own boundary.
Keep inference optimization enabled for most apps. Start with an account-level monthly cap, then add closer connection-level limits for agents, shared tools, or flows with separate budgets.
Inference optimization
Inference optimization is deliberate model choice. It lets Ando select the model route suited to the task while keeping usage visible under the same connection. The goal is to match capability to the moment without making every app choose a model by hand.
Over time, Ando can show which apps are using more context, which sessions ask for longer responses, and where a connection may need a closer boundary.
Ando can choose an appropriate model route for the request while keeping the connection clear.
Optimization weighs capability and usage so everyday requests can stay appropriately routed.
Requests stay attached to the app and connection that generated them, making review simple.
The app keeps the same credential and account boundary even when Ando adjusts routing.
Spend controls
Spend controls let users set inference token caps at the account level. That includes an overall monthly cap, plus lower limits where the product supports a narrower connection or flow boundary.
Use account-level caps to protect the total budget. Use connection-level controls when an app, agent, or shared tool should have a closer operating limit.
Set the monthly account boundary first so all app usage has a ceiling.
Add narrower limits for agents, shared tools, and experiments.
Review token usage, expensive routes, and repeated flows before changing caps.
When to adjust settings
Adjust settings when an app's behavior changes, when a new flow is introduced, or when usage rises. Tighten connection-level caps before broad account caps if one flow is responsible for the change. Keep optimization enabled unless repeatability, evaluation, or compatibility requires a fixed model path.