Local Memory Service 'memnest' Released to Prevent AI Coding Agents from Losing Session Context

The local memory service memnest has been released to address the issue of AI coding agents (such as Claude Code, Codex, and pi) forgetting previous conversations or decisions when sessions change. This tool processes data exclusively on the user's machine and adopts a method of storing and retrieving original text without LLM summarization.

Key Features and Architecture

• Persistent Memory Management: Maintains decisions, preferences, modifications, and rules across sessions. Memories are separated by directory (workspace), but global rules (playbooks) can be shared.
• Hybrid Search: Combines BM25 keyword matching with multilingual vector similarity to enhance search accuracy. The embedding model runs locally, with the default model occupying approximately 1.1GB on disk.
• Security and Data Management: Credentials are stored in an AES-256-GCM encrypted vault, separate from searchable memories. The database uses SQLite, and the search index is rebuildable.
• Automatic Context Capture: Via memnest hook, relevant context can be automatically injected before Claude Code or Codex prompts. Additionally, memnest watch monitors and saves conversation logs.

Integration and Usage

It operates as a single Rust-based engine and integrates with MCP (Model Context Protocol) clients via HTTP or Stdio. On Linux (aarch64/x86_64), binary installation is possible without the Rust toolchain. Developers can manage memories through five standard tools, such as memory_remember and memory_search, and control the search scope per project.