GPU Services

IntelligenceBox includes specialized GPU services for AI-intensive operations. MCP servers can call these services to perform document analysis, embeddings generation, entity extraction, and more.

Available Services

Accessing from MCP Servers

GPU services run on the internal Docker network. From your MCP server container, access them at:

http://<service-name>:<port>

Example:

// From your MCP server
const response = await fetch("http://colpali-server:8001/embed", {
  method: "POST",
  headers: { "Content-Type": "application/json" },
  body: JSON.stringify({ images: [imageBase64] }),
});

Common Patterns

Document Processing Pipeline

1. Extract text → Docling (/process)
2. Generate embeddings → ColPali (/embed) or Jina Embeddings (/embeddings/text)
3. Extract entities → GLiNER (/predict)
4. Extract tables → Table Extraction (/extract-tables)

Search and Retrieval

1. Query embedding → Jina Embeddings (/embeddings/text)
2. Sparse vectors → FastEmbed (BM25/SPLADE)
3. Rerank results → Jina Reranker (/rerank)