Glossary and troubleshooting
Use these definitions while reading the code. Follow the reference links for implementation details.
Terms in the code
Section titled “Terms in the code”| Term | In this application |
|---|---|
| Model | The Foundry-deployed model that proposes text and tool calls |
| Model deployment | A named model instance the client calls, such as the workshop’s chat deployment |
| Agent | Instructions, model access, context, and tools run by Agent Framework |
| Instructions | Rules that guide the agent’s behavior during a run; application code must enforce important constraints |
| Tool | Application code exposed through a typed, described contract |
| Tool schema | Metadata describing a tool’s name and inputs; the model does not receive its C# implementation |
| Tool call | A model-requested invocation with arguments; inspect the result to know whether it succeeded |
| MCP | The protocol used to discover and call tools across services |
| AG-UI | The protocol between the WebUI server’s AGUIChatClient and the agent’s /ag-ui endpoint |
| Blazor circuit | The server-side UI connection and component state associated with a browser connection |
| Handoff | A transfer of control between specialists along permitted edges |
| Workflow | The topology and orchestration of agent execution |
| Conversation context | Messages and state available to an agent run |
| AgentSession | Framework conversation state passed between agent runs; persistence depends on the agent and storage design |
| Interview record | The persisted business data in Cosmos: inputs, transcript, completion |
| Aspire | Local/service orchestration, resource references, health, and telemetry plumbing |
| Foundry | The Azure model backend used in the core path |
| Evaluation | Comparing observed agent behavior with expected behavior across representative inputs and repeated runs |
| WorkshopHosting | Supplied model/authentication and DevUI helper activated in the first-agent lesson |
Find the failed boundary
Section titled “Find the failed boundary”| Symptom | First evidence to inspect |
|---|---|
| Nothing starts | CLI/SDK output and AppHost startup |
| Cosmos or parser is unhealthy | Container runtime and the specific resource logs |
| Model authentication failure | Identity, tenant, token scope, and resource permissions |
| No tool calls | Agent tool registration and discovered tool list |
| “Saved” but no record | InterviewData result and database contents |
| Duplicate transcript | New text passed to the append operation |
| UI stream parsing error | Correct /ag-ui endpoint and version-specific serialization adapter |
| Refresh starts a new chat | Blazor circuit and session lifecycle |
The troubleshooting reference covers the full failure list. For a record or refresh question, use state ownership and update semantics. For routing, use the handoff graph.
Foundry can also host agent code through its managed agent runtime. That is a separate architecture from this workshop’s local .NET agents calling a Foundry model. See hosting and extensions before changing the runtime.
Share the checkpoint, failing operation, and redacted error. Keep credentials, document contents, and transcripts private.