AI

What is AI?

AI provides tools for artificial intelligence operating on top of resource sets – collections of interconnected models. It includes “Narrator” processors that describe model elements and their relationships in multiple ways, and supports embeddings, vector stores, and chat completions.

How to use AI?

Use the CLI commands: a vector store mixin, an embeddings generator for search-documents.json, a vector store generator (with or without embeddings, new or add/update existing), and semantic search HTTP server routes. A Chat Vue.js component is adapted for chat interfaces.

Key features of AI

• Narrator processors describe model elements and relationships.
• Supports OpenAI and Ollama for embeddings.
• Uses hnswlib vector store for semantic search.
• CLI commands for generating embeddings and vector stores.
• Semantic search HTTP server routes.
• Chat components for static site or model interaction.

Use cases of AI

• Explaining familial relationships (e.g., parent, father, sibling) from a model.
• Generating descriptions with narration and storing them as vector embeddings.
• Performing semantic search that considers both semantic and graph distance.
• Building a chat interface for a static site using OpenAI or Ollama.

FAQ from AI

What embeddings providers are supported?

OpenAI and Ollama are explicitly mentioned for embeddings generation.

What vector store library is used?

The server uses hnswlib (from https://github.com/jelmerk/hnswlib). Observations include 10 minutes to load 200K vectors and ~2.5 Gb index file.

What chat capabilities are included?

Chat completions are supported. A CLI enables chatting with a site (serving static site + semantic search) or chatting with a model using OpenAI or Ollama on a GCP VM with Docker.