Files
2026-07-13 12:58:18 +08:00

92 lines
3.9 KiB
C#

using Microsoft.Agents.AI;
using Microsoft.Extensions.AI;
using OpenAI;
/// <summary>
/// Factory for the Open-Ended Generative UI (minimal) demo agent.
///
/// The simplest possible example that exercises the open-ended generative
/// UI pipeline. All the interesting work happens outside the agent:
///
/// - The CopilotKit runtime's <c>openGenerativeUI</c> flag (see
/// <c>src/app/api/copilotkit-ogui/route.ts</c>) auto-injects the
/// frontend-registered <c>generateSandboxedUi</c> tool. The LLM sees it
/// via the normal AG-UI flow.
/// - When the LLM calls <c>generateSandboxedUi</c>, the runtime's
/// <c>OpenGenerativeUIMiddleware</c> converts the streaming tool call
/// into <c>open-generative-ui</c> activity events that the built-in
/// renderer mounts inside a sandboxed iframe.
///
/// This is the minimal variant: no sandbox functions, no app-side tools.
/// The agent simply asks the LLM to design and emit a single-shot
/// sandboxed UI. The "advanced" sibling (<see cref="OpenGenUiAdvancedAgentFactory"/>)
/// builds on this with sandbox-to-host function calling via
/// <c>openGenerativeUI.sandboxFunctions</c>.
/// </summary>
public class OpenGenUiAgentFactory
{
private const int HarnessMaxContextWindowTokens = 128_000;
private const int HarnessMaxOutputTokens = 8_192;
private const string SystemPrompt = @"You are a UI-generating assistant for an Open Generative UI demo
focused on intricate, educational visualisations (3D axes / rotations,
neural-network activations, sorting-algorithm walkthroughs, Fourier
series, wave interference, planetary orbits, etc.).
On every user turn you MUST call the `generateSandboxedUi` frontend tool
exactly once. Design a visually polished, self-contained HTML + CSS +
SVG widget that *teaches* the requested concept.
The frontend injects a detailed ""design skill"" as agent context
describing the palette, typography, labelling, and motion conventions
expected — follow it closely. Key invariants:
- Use inline SVG (or <canvas>) for geometric content, not stacks of <div>s.
- Every axis is labelled; every colour-coded series has a legend.
- Prefer CSS @keyframes / transitions over setInterval; loop cyclical
concepts with animation-iteration-count: infinite.
- Motion must teach — animate the actual step of the concept, not decoration.
- No fetch / XHR / localStorage — the sandbox has no same-origin access.
Output order:
- `initialHeight` (typically 480-560 for visualisations) first.
- A short `placeholderMessages` array (2-3 lines describing the build).
- `css` (complete).
- `html` (streams live — keep it tidy). CDN <script> tags for Chart.js /
D3 / etc. go inside the html.
Keep your own chat message brief (1 sentence) — the real output is the
rendered visualisation.";
private readonly OpenAIClient _openAiClient;
public OpenGenUiAgentFactory(OpenAIClient openAiClient)
{
ArgumentNullException.ThrowIfNull(openAiClient);
_openAiClient = openAiClient;
}
public AIAgent CreateAgent()
{
var chatClient = _openAiClient.GetChatClient("gpt-4o-mini").AsIChatClient();
// No backend tools. The `generateSandboxedUi` tool is registered
// on the frontend by CopilotKitProvider (when `openGenerativeUI`
// is enabled on the runtime) and merged into the agent's tool
// list by the AG-UI protocol as a frontend-side action.
return chatClient.AsHarnessAgent(
HarnessMaxContextWindowTokens,
HarnessMaxOutputTokens,
new HarnessAgentOptions
{
Name = "OpenGenUiAgent",
Description = "Open Generative UI (minimal) powered by Microsoft Agent Harness over Microsoft Agent Framework.",
ChatOptions = new ChatOptions
{
Instructions = SystemPrompt,
MaxOutputTokens = HarnessMaxOutputTokens,
},
});
}
}