Skip to content
Arbitex Docs

Models & Routing

Models & Routing is where you connect AI providers, browse the model catalog, govern which models are approved for use, monitor your outpost fleet, and register the servers and sources your agents can reach — MCP servers and RAG retrieval sources.

The Provider Credentials panel in the Models & Routing group of the admin console, showing the Add credential form and a table of stored credentials with no secret ever displayed.
Admin console → Models & Routing → Provider CredentialsCapture pending — Citadel synthetic data on a non-production Arbitex demo instance. Never client data.

What’s in this group

Section titled “What’s in this group”

Each panel below maps 1:1 to a tab in the console’s Models & Routing group, in the order they appear in the sidebar.

Agent Tool PoliciesAuthor allow/deny tool-use policies for agents — under development, no effect until the agentic-AI scanning model ships.
Fleet & OutpostMonitor outpost health, certificate expiry, connectivity, and site grouping.
Image OCRChoose the OCR engine and language for extracting text from images before DLP inspection.
MCP ServersRegister MCP servers, author per-tool allow/deny policy, and dry-run policy decisions.
Model CatalogBrowse the read-only base model catalog and set org-wide and per-group model access.
Model Risk ManagementAssign risk tiers and validation status, and run the approve/reject governance workflow.
Provider CredentialsStore BYO provider API credentials as envelope ciphertext — the secret is never displayed or returned.
Retrieval Sources (RAG)Register vector database, web crawler, S3, Confluence, and SharePoint sources for grounding.
Section titled “Related”

Routing and the model kill switch are not configured in this group. The standalone routing surface was retired — routing is now the pinned, route-only pack at chain position [0], authored in Security & DLP → Chain alongside your DLP policy packs. The kill switch is likewise authoritative in Security & DLP → Global Kill Switch, covering global, provider, model, and feature scope — there is no separate per-model kill switch here.