Skip to content

ai-integration.md

Latest commit

 

History

History
107 lines (71 loc) · 2.78 KB

ai-integration.md

File metadata and controls

107 lines (71 loc) · 2.78 KB

AI Integration

uberSKILLS uses OpenRouter as its AI provider, accessed through the Vercel AI SDK with the @openrouter/ai-sdk-provider bridge.

Overview

Client (useChat)  -->  API Route (streamText)  -->  OpenRouter  -->  AI Model
                  <--  SSE stream              <--             <--

All AI calls are server-side. The OpenRouter API key is decrypted from the database only in API route handlers and never exposed to the client.

OpenRouter

API Key

  • Stored encrypted (AES-256-GCM) in the settings table.
  • Decrypted server-side in API route handlers.
  • Never logged, never included in error messages, never sent to the client.

Models Endpoint

GET https://openrouter.ai/api/v1/models

Used to populate model selector dropdowns. Results are cached in the database.

Chat Completions

POST https://openrouter.ai/api/v1/chat/completions

Used for skill creation and testing. Responses are streamed via SSE. Token usage is returned in the final chunk.

Required Headers

All requests to OpenRouter include:

HTTP-Referer: http://localhost:3000
X-Title: uberSKILLS

Vercel AI SDK

Server-Side -- streamText()

Used in /api/chat and /api/test route handlers:

import { streamText } from "ai";
import { createOpenRouter } from "@openrouter/ai-sdk-provider";

export async function POST(req: Request) {
  const { messages, model } = await req.json();
  const apiKey = await getDecryptedApiKey();

  const openrouter = createOpenRouter({ apiKey });

  const result = streamText({
    model: openrouter(model),
    system: systemPrompt,
    messages,
  });

  return result.toDataStreamResponse();
}

Client-Side -- useChat()

Used in React components for chat and testing interfaces:

import { useChat } from "ai/react";

const { messages, input, handleInputChange, handleSubmit, isLoading } = useChat({
  api: "/api/chat",
  body: { model: selectedModel },
});

Key Features Used

Feature Usage
streamText() Server-side streaming in API routes
useChat() Client-side chat hook with automatic streaming
toDataStreamResponse() Convert stream to Next.js-compatible Response
onFinish callback Capture token usage and save to test_runs

System Prompts

Skill Creation

When using AI-assisted creation, the system prompt instructs the model to generate valid SKILL.md output with:

  • YAML frontmatter containing name, description, and trigger fields
  • Clear, actionable markdown instructions
  • Example trigger scenarios

Skill Testing

The resolved skill content itself serves as the system prompt. Any $ARGUMENTS or $VARIABLE_NAME placeholders are substituted with user-provided values before sending.