Built different
FlakeCache isn't just another cache. It's a universal build artifact store with revolutionary deduplication.
VectorCDC: 4x better deduplication
Traditional caches store files as blobs. FlakeCache uses VectorCDC (Vectorized Content-Defined Chunking) to split files using SIMD instructions (AVX2/NEON).
Result? Blazing fast chunking that finds 4x more duplicates than standard methods. Node 20 shares 90% of chunks with Node 18.
- • Industry standard: 1.48x dedup ratio
- • 80% of files are duplicates (but undetected)
- • VectorCDC approach: 4x+ dedup ratio
# Push Docker image
docker push cache.flakecache.com/myorg/app:v1
# Chunking layer...
Layer 3: 200 MB → 40 chunks
✓ 35 chunks cached (from Nix!)
✓ 5 new chunks uploaded
Uploaded: 12 MB (94% dedup)Docker Registry v2
Full Docker Registry v2 API support. Push and pull images with standard Docker commands. No proxies, no wrappers.
The magic? Docker layers are chunked with the same FastCDC algorithm. Your Docker images automatically share chunks with your Nix packages.
Cross-language deduplication
Build a Spring Boot app with Maven. The JAR is chunked and stored. Now build a Docker image with that JAR.
FlakeCache recognizes the chunks are identical. Zero additional storage. This works across Nix, Docker, Maven, Cargo, and more.
Enterprise features
Built for teams that need more
SSO / SAML
Integrate with Okta, Azure AD, Google Workspace
Audit Logging
Full audit trail for compliance requirements
Multi-Tenant
Organizations with granular access control
GDPR Ready
Data residency and privacy compliance
Research-proven, finally shipped
Academic papers have proven FastCDC deduplication for Docker registries since 2020. Nobody shipped it commercially. Until now.
DupHunter (USENIX ATC 2020) and ACM Transactions on Storage 2024 proved these techniques work. We're the first to ship them in a production registry with cross-language deduplication.