AI Setup is the workspace-wide configuration interface for AI models, customization settings, guardrails, and people directory management — every agent in the project inherits it.
Overview
AI setup — labeled AI Agent Settings — is the "Configuration interface for AI models, customization settings, guardrails and people directory management." Settings here apply to every agent in the project, not to one agent, and a new agent inherits these defaults when it's created. The settings forms (Agent Workflow, Guardrails and Knowledge) save automatically — there's no Save button. The People Schema and the spending limits and alert rules have their own Save buttons. Use AI setup to set sensible org-wide defaults once, so individual agents start from a good baseline.
What's in AI Setup
The left nav groups the settings:
Section | Covers |
|---|---|
Models | Browse the workspace's model catalog (the Model Library). |
Agent Workflow | The default AI Models (per-function model pickers), Indexing, Pre Processing, and Response Generation that new agents inherit. |
Spending | Workspace-wide Usage, Limits, and Alerts for AI cost. |
Guardrails | Default PII Masking, Hallucinations, Content Filters, Custom Word Filter, Denied Topics, and Blocked Messaging. |
People | A directory of people the agents can know about, synced from source apps. |
Knowledge | A master Enrichment Schema for knowledge chunks, plus PII masking for knowledge. |
Who Can See and Change AI Setup
What you see in AI setup depends on your permissions:
To change the settings forms, you need edit permission on AI Agent Settings. With view permission only, every form is read-only.
Models, Limits, Alerts and People appear in the left nav only if you have the matching permission.
Until a people source is connected, People shows only Sources.
How Workspace Defaults Relate to a Single Agent
The AI Models, Indexing, Pre Processing, Response Generation, and Guardrails here are the same forms you'd configure on an individual agent — but set at the workspace level. They define the defaults a new agent starts from; an agent can then override them on its own screens. Configure the common baseline once in AI setup, and only adjust per-agent where an agent genuinely needs something different.
Where to Add Model Provider API Keys
Not in these forms. The model pickers in AI setup choose from a managed model catalog — the models added in the Model Library — rather than taking raw provider API keys. Provider credentials and connections are managed where models are registered, not on the setup settings forms. Pick defaults here; manage the models and their connections in the Model Library.