Changelog
All notable changes to Recallm will be documented in this file.
The format is based on Keep a Changelog.
[0.1.0] — 2026-03-06
Added
SemanticCache.wrap()with sync and async supportInMemoryStorage— zero-dependency in-process backendRedisStorage— persistent backend with lazy tombstone cleanupFastEmbedEmbedder— ONNX-based, ~20MB default embedderSentenceTransformerEmbedder— optional torch-based embedder- Three similarity threshold profiles:
strict(0.97),balanced(0.92),loose(0.85) - Namespace-based cache invalidation
- TTL support on cache entries
- Prometheus metrics: hits, misses, errors, embedding latency, similarity scores, stream bypass
- Structlog structured logging on all cache events with rich operational fields
stream=Truebypass with per-namespace counter- Fail-open on all cache operation failures
CacheContexttype alias for type-checking convenienceSemanticCache.async_warmup()for non-blocking model load in async frameworks- Grafana dashboard (8 panels,
$namespacevariable) - Benchmark suite with four realistic prompt distributions