v0.1.0

v0.1.0 - First release on PyPI

Persistent context, semantic recall and the vector-store contract, for Python. The Python port of particle-academy/prism-memory. The distribution is prism-ai-memory; the package imports as prism_memory.

pip install prism-ai-memory

What it has

• Vector.of() validates every component when the vector is built. An empty vector, a non-number, NaN or infinity, a zero vector, or one too large to score is refused with invalid_vector.
• The storage format is base64 of little-endian 64-bit floats, byte for byte what the PHP reference writes, so PHP, Python and TypeScript applications can read each other's rows.
• recall() over-fetches candidates by similarity, then rescores them with a Weighting of relevance and exponential recency decay. Every similarity returned is an exact cosine.
• BinarySignature: a deterministic bit signature for picking candidates without reading full vectors.
• The VectorStore protocol, and InMemoryVectorStore, which reports itself volatile and refuses to mix two embedding models in one collection.
• Every failure is a MemoryError with a stable code.
• No runtime dependencies. Python 3.10 or later.

The package does not call an embedding model; pass the vectors your provider returns.

prism-parity's memory-vector-storage corpus pins the storage format and scoring against the PHP reference and the TypeScript port.