Single model call vs prompt agent vs hosted agent vs Toolbox vs Models catalog

Verdict: No tools, no orchestration: a single model call. Instructions, a model and tools set through the portal with no code to maintain: a prompt agent. Your own containerised orchestration code: a hosted agent. Comparing providers first: the Models catalog.

CriterionSingle model callPrompt agentHosted agentToolboxModels catalog
What it isPrompt-in, completion-out, no agent involvedInstructions, a model and tools specified in the portal or SDKYour own code or container, e.g. LangGraph, packaged as a source package or container imageCurates tools, memory and retrieval for an existing agentBrowsable list of over ten thousand models from providers including Microsoft, OpenAI, Anthropic and Meta
Code or containers to maintainNoneNone - Foundry hosts and runs itYes - the team's own runtime code is what Foundry containerisesNone on its own; extends an agentNone; a catalog, not a runtime
Identity modelN/AHas an Entra agent identity, but not one automatically dedicated per agentEach hosted agent automatically receives its own dedicated Microsoft Entra identityN/AN/A
ScalingN/AManaged by FoundryScales by adding container instances per session and request volumeN/AN/A
Choose whenA plain completion with no tools or retrievalPrototyping quickly with no dedicated platform engineeringExisting custom orchestration code must run as a managed, autoscaling endpointAn existing agent needs extra tools or retrieval attachedScoping a generative AI workload and comparing candidate models first

Rules

Traps