AI Features — ApexKit Docs

# AI Features ApexKit is built from the ground up to be **AI-Native**. It integrates machine learning capabilities directly into the core engine. ## 1. AI Architect The AI Architect is your co-pilot for building backends. Accessible from the dashboard, it can: - **Design Schemas**: Describe your app, and it will create the collections and fields. - **Write Scripts**: Ask it to write a specific Edge Function. - **Explain Logic**: Highlight a policy and ask "How does this work?". ## 2. Vector Semantic Search Standard search looks for keywords. **Vector search** looks for meaning. ### How it works: 1. Enable `Vectorize` on a text field in your collection. 2. ApexKit automatically generates high-dimensional embeddings using local models (e.g., `BGE`). 3. Use the SDK to find relevant records. ```javascript // Search for "healthy recipes" even if the text says "nutritious meals" const results = await apex.collection('recipes').searchTextVector("healthy recipes", 10); ``` ## 3. AI Actions (Prompts) AI Actions are server-side prompt templates that you can expose as API endpoints. ### Defining an Action: - **Slug**: `summarize-article` - **System Prompt**: "You are a helpful assistant that summarizes long text." - **Prompt Template**: "Please summarize this: {{input}}" ### Executing an Action: ```javascript const res = await apex.ai.run('summarize-article', { input: "A very long text about technology..." }); console.log(res.result); ``` ## 4. Local Embeddings & Models ApexKit uses the **Candle** library to run inference locally whenever possible. This means: - **Privacy**: Your data doesn't always have to leave your server. - **Speed**: No network latency for embedding generation. - **Cost**: No per-token pricing for vector generation. ## 5. Instant AI Search ApexKit combines **Tantivy** (keyword search) with **Vector search** to provide "Hybrid Search" results that are both accurate and fast. ### Instant Search Example (WebSocket): ```javascript const results = await realtime.search('products', 'red sneakers', 5); ``` This returns results in milliseconds as the user types, using the combined power of full-text and semantic indexing.