Show HN: Open-source agent memory layer, 96% on LongMemEval, local-first (github.com)

Awareness Local has been introduced as an innovative local-first memory server designed specifically for AI coding agents, enabling them to maintain persistent memory across sessions without requiring cloud accounts. With a one-command setup, developers can integrate this solution into compatible IDEs such as Cursor, Claude Code, and Copilot. Awareness Local leverages a hybrid semantic and keyword retrieval system, storing memory in Markdown format and indexing it with SQLite FTS5 for efficient access and recall. Evaluated using the LongMemEval benchmark, it achieved an impressive 96% recall rate at the R@5 level, demonstrating its effectiveness in long-term contextual memory. The significance of Awareness Local lies in its ability to address "lobster memory"—the phenomenon where AI agents lose context between interactions. This technology allows agents to remember architectural decisions, pending tasks, and implementation details, streamlining workflows for developers. The standalone, offline functionality also ensures privacy and flexibility, making it suitable for various scenarios such as codebase migrations and team collaboration. By enhancing the memory capabilities of AI agents, Awareness Local sets a new standard for improving productivity and continuity in software development projects.