Start with our MIT-licensed core engine, scale with enterprise-grade support and hardware integration
slimRAG uses end-to-end ChaCha20-Poly1305 encryption. Data is encrypted at the edge (slimSync) and decrypted only at the RAG server (slimRagSvr). The relay hub (slimHub) stores only ciphertext and blind-IDs — it is zero-knowledge by design and cannot decrypt any payload.
The core engine runs on any modern x86_64 or ARM64 Linux system with Rust 1.75+. For CPU-only deployments, FastEmbed (ONNX) uses the CPU. For GPU acceleration, the LlamaCpp backend supports CUDA and Metal. For maximum performance, the IronCurtain NPU adaptation provides hardware-optimized inference.
Yes. slimRAG indexes code via Tree-sitter AST parsing (symbol dependency graph) and documents via FastCDC content-defined chunking with paragraph boundary detection. Both are stored in the same LMDB index and searched via the hybrid retrieval engine.
Traditional cloud RAG sends your data to third-party servers for indexing and inference. slimRAG runs entirely on your infrastructure — data never leaves your premises. This eliminates data exfiltration risk, ensures regulatory compliance (GDPR, HIPAA, SOC2), and removes dependency on cloud network latency.