Show HN: Yet Another Memory System for LLM's
10 days ago
- #search
- #LLM
- #storage
- Persistent memory for LLMs with content-addressed storage, deduplication, and semantic search.
- Features include SHA-256 based storage, block-level deduplication, Zstandard/LZMA compression, and full-text/semantic search.
- Supports crash recovery via write-ahead logging and achieves high performance (100MB/s+ throughput).
- Available on Linux x86_64, ARM64, and macOS (Intel/Apple Silicon).
- Docker and Homebrew installation options provided.
- Requires C++20 compiler, CMake 3.20+, and Python 3.8+ for Conan builds.
- Includes CLI with TUI browser, MCP server, and support for PDF text extraction.
- Designed for seamless integration with LLMs, supporting storage, retrieval, and search operations.
- Optimization options for large files and memory usage adjustments available.
- Licensed under Apache-2.0.