Normalization & Decomposition
Normalize input and identify atomic information units suitable for durable memory.
The Librarian is the intelligence and orchestration layer that transforms unstructured AI interactions into durable, structured memory. It coordinates normalization, semantic decomposition, extraction, classification, provenance, contradiction analysis, and persistence around a model-independent memory substrate.
Ollama, Mistral, Qwen, Llama, hosted inference APIs, and future UNIMATRIX model substrates are inference providers—not the Librarian. They supply language inference to stages that are governed by the UNIMATRIX orchestration architecture.
Normalize input and identify atomic information units suitable for durable memory.
Identify entities and relationship candidates for structured lineage and graph representation.
Assign semantic categories and evidence-aware confidence signals to extracted information.
Compare new assertions with existing memory and identify potential conflicts and supersession relationships.
Preserve when information was recorded, its validity context, and its source lineage.
Persist structured memory and expose it through governed semantic, keyword, and graph-oriented retrieval.
UNIMATRIX distinguishes implemented, experimental, and planned capabilities. Current server code includes normalization, semantic chunking, extraction, classification, contradiction checks, evidence-aware retention, and source lineage. The 2.0 integration also links indexed memories into an Associative Memory Graph and connects worker consolidation to higher-level context. These mechanisms require matching migrations and configured services; their presence in code does not establish extraction accuracy, calibrated truth, or hosted availability.
Review the 2.0 architecture →