HN Showcase
Back to browse
CacheCore – semantic agent caching with dependency invalidation

CacheCore – semantic agent caching with dependency invalidation

by fabriziorocco·May 7, 2026·2 points·4 comments
Visit ProjectView on HN

AI Analysis

●●●BangerBig BrainSolve My Problem

Semantic caching with dependency invalidation beats standard Redis wrappers for agent costs.

Strengths
  • •Two-tier caching architecture returns exact matches in 70ms and semantic matches under one second.
  • •Automatic dependency invalidation handles context changes without manual cache busting logic.
  • •Drop-in proxy configuration requires only a base_url change with no SDK modifications.
Weaknesses
  • •Request-access gate for early testers limits immediate verification of the claimed performance metrics.
  • •Semantic matching accuracy depends heavily on embedding quality which varies by domain.
Category
AI/ML
Target Audience

Engineers building multi-agent systems or RAG applications

Similar To

Portkey · LiteLLM · Redis

Post Description

I want to present you a project I built over the last few weeks. It's called CacheCore and it is a semantic cache specifically for RAGs and Agents.

It works as proxy by replacing the traditional base_url with the CacheCore gateway.

Gateway is built in Rust and there are 2 levels of caching (L1 and L2). There is a automatic mechanism for dependency invalidation.

Compared to other tools: most solutions require you to change the code logic and don't have dependency invalidation.

I started testing on the Legal and Health domains where context reuse is high and query costs add up fast.

Similar Projects