The Admin Console
The admin console is where operators and developers configure, manage, and monitor AI agents. It’s a separate interface from the Hub — designed for building and governance, not end-user chat.
Dashboard
The admin console opens to a dashboard showing your workspace and all apps (agents) within it. From here you can:
- See all configured agents at a glance
- Create new apps
- Navigate to any app’s detail view
- Access instance-level settings (models, configuration)
App Detail View
Each app has a tabbed interface:
Build & Test
A debug chat interface. You interact with the agent and see exactly what’s happening:
- Which model is being called
- What tools the agent invokes and their responses
- The full reasoning trace
- File system operations
This is the test environment — changes to intelligence, connections, and channels take effect here immediately.
Intelligence
Configure the agent’s brain:
- Intelligence type — Dynamic Prompt (production) or Chat Completion
- Model selection — choose from enabled models
- System prompt — define the agent’s personality, role, and constraints
- See Intelligence Configuration → for details
Connections
Add and manage tool integrations:
- MCP — connect to MCP servers that expose tools to the agent
- OpenAPI — connect to APIs via OpenAPI specifications
- Each connection can be enabled or disabled per app
- See Tools: MCP → and Tools: OpenAPI →
Channels (Reach)
Configure where the app is accessible:
- Teams — enable Teams integration, download the app manifest
- Web Chat Widget — configure and generate the embed snippet
- DirectLine — always enabled (powers the Hub)
- See Channels →
Analytics
Monitor agent behavior and usage:
- Conversations — total count, message counts, duration
- LLM usage — token consumption by model
- Tool calls — which tools are being called, how often, success rates
- Usage patterns — hourly and daily activity trends
Activities
Browse individual conversation turns — see what happened in each interaction, what tools were called, and how the agent responded.
Versions
Publish and manage versions:
- Publish a new version to make current test configuration live
- Browse version history
- See Publishing & Versions →
Instance-Level Settings
Beyond individual apps, the admin console provides instance-wide management:
Models
Navigate to Instance → Models to:
- Sync available models from your AI Foundry catalog
- Enable or disable models across the instance
- See which models are deployed and their status
Agent Flow Diagram
Each app has a visual agent flow diagram (built with @xyflow/svelte) that shows how the agent is composed — its intelligence type, tools, connections, and sub-agent relationships.
This section is a work in progress.
Permissions
Access to the admin console requires the Management global role, assigned via Entra ID app role assignment. Within the console, individual app operations are governed by per-node permissions (Read, ReadAndWrite, ReadAndPublish, etc.).