schwabauerbriantomas-gif/m2m-vector-search

Implements probabilistic Gaussian Splat vectors (μ, κ, α parameters) with Energy-Based Models for uncertainty quantification in similarity scores, enabling confidence-aware retrieval for semantic memory and AI agent applications. Supports CPU, CUDA, and Vulkan compute backends through a unified API, with hybrid search combining vector similarity and BM25 keyword matching via Reciprocal Rank Fusion. Includes temporal decay for memory recency weighting, self-organized criticality consolidation for long-term storage management, and a FastAPI REST endpoint for distributed edge/coordinator cluster deployments.

Available on PyPI.