Why should you care?

Long ERP narratives and clinical summaries feel broken at 8-second waits; streaming shows progress and allows early cancel.

See it live — copy this example

Paste into an ASP.NET Core 8+ / AgentNest project, then run with dotnet run (set your API keys in user-secrets).

// AgentNest.Api/Endpoints/StreamEndpoints.cs
app.MapGet("/api/crm/stream/{accountId:guid}", async (
    Guid accountId, ICrmCopilotAgent agent, HttpContext ctx, CancellationToken ct) =>
{
    ctx.Response.Headers.ContentType = "text/event-stream";
    await foreach (var chunk in agent.StreamAccountBriefAsync(accountId, ct))
    {
        await ctx.Response.WriteAsync($"data: {chunk}\n\n", ct);
        await ctx.Response.Body.FlushAsync(ct);
    }
});

What happened?

• The endpoint sets text/event-stream and writes SSE data lines per chunk.
• CancellationToken aborts OpenAI streams when the browser closes the connection.

Practice next

1. Add StreamAccountBriefAsync returning IAsyncEnumerable on the agent.
2. Use chat.GetStreamingResponseAsync from Microsoft.Extensions.AI.
3. Test with curl -N to see incremental output.
4. Wrap chunks as JSON { "delta": "..." } for typed front-end parsers.
5. Add heartbeat comments every 15s for load balancers.

Remember

Stream with IAsyncEnumerable or SSE for AgentNest chat UIs. Propagate RequestAborted to cancel model generation. Set text/event-stream and flush after each chunk.

CRM live brief

Reps watch account summaries appear word-by-word before a call.

Outcome: Streaming cuts perceived latency and lets users cancel off-topic generations.