Integrating models

Models are the AI engines that power AIDB’s core operations — embedding generation, text chunking, summarization, OCR, and completions. Every pipeline step and SQL function that performs an AI operation requires a model to be registered and reachable before it can run.

When to add models: Register your models after installing and configuring AIDB, and before creating pipelines or knowledge bases that reference them. If you’re using one of AIDB’s built-in local models (such as BERT or CLIP), no registration is needed — they’re available immediately. For external APIs (OpenAI, NVIDIA NIM, Gemini) or custom HuggingFace variants, use aidb.create_model() to register the model first.

Page	What it covers
Local models	Built-in local models (BERT, CLIP, T5, Llama) and how to register custom variants
Support matrix	Pre-vetted HuggingFace model variants confirmed to work with AIDB's local providers
External model connections	Connecting to OpenAI-compatible APIs, NVIDIA NIM, Google Gemini, and OpenRouter

• Local models
  • Built-in models
  • Custom local models
    • BERT
    • CLIP
    • T5
    • Llama
  • Using a registered model
• Support matrix
  • Local embedded models
    • BERT
    • CLIP
    • Llama
    • T5
  • NIM models
    • NIM completions (nim_completions )
    • NIM embeddings (nim_embeddings )
    • NIM CLIP (nim_clip )
    • NIM OCR (nim_paddle_ocr )
    • NIM reranking (nim_reranking )
• External model connections
  • OpenAI-compatible endpoints
    • Embeddings
    • Chat completions
  • NVIDIA NIM
    • Text generation (NIM completions)
    • Multimodal embeddings (NIM CLIP)
    • OCR (NIM OCR)
    • Reranking (NIM reranking)
  • Google Gemini
  • OpenRouter
    • Chat completions
    • Embeddings