Developer & Model Specs

LLM Schemas, MCP & Retrieval Protocols

Tensor Knowledge catalogs are delivered in uniform, machine-consumable formats designed for immediate consumption by modern reasoning models, agentic runtimes, and vector stores.

Zero Friction Integration: Compatible out-of-the-box with OpenAI Function Calling, Anthropic Claude Tools, Model Context Protocol (MCP), LangChain, and LlamaIndex.

Distribution Protocols

Protocol

Model Context Protocol (MCP)

Pre-packaged MCP servers allowing desktop and server AI clients to query catalog endpoints directly via stdio or SSE transport.

Schema

JSON-LD & RDF Ontologies

Linked data standards providing strict semantic types, parent-child relationships, and machine-verifiable assertions.

Retrieval

Vector & Hybrid Chunk Packs

Context chunks optimized for cosine similarity and BM25 keywords, paired with pre-calculated metadata filters for instant Qdrant/Pinecone/Milvus loading.

Benchmarking

Eval Harness & Golden Testsets

Curated question-context-answer triples for running Ragas, TruLens, and DeepEval test suites against your retrieval pipeline.

OpenAPI / Function Calling Definition Example

Standardized tool schema format for agentic orchestration:

{
  "name": "tensor_knowledge_query",
  "description": "Query verified domain knowledge catalogs for factual context, entity relations, and rules.",
  "parameters": {
    "type": "object",
    "properties": {
      "catalog": {
        "type": "string",
        "enum": ["travel", "auto", "learning"],
        "description": "Target domain catalog"
      },
      "query": {
        "type": "string",
        "description": "Semantic lookup or entity identifier"
      },
      "format": {
        "type": "string",
        "enum": ["summary_chunk", "entity_graph", "json_ld"],
        "default": "summary_chunk"
      }
    },
    "required": ["catalog", "query"]
  }
}

Coming soon

NPM: npx @tensorknowledge/mcp-server start
Python: pip install tensorknowledge
Direct Download: Download standalone Parquet or JSONL dumps for offline airgapped LLM deployments