Open source, community driven memory plugin for Codex CLI & Openclaw

A community-driven project called Signet is tackling the noise in AI memory systems with a transparent, open-source solution. Built over two months by a group of developers, Signet is a memory plugin designed for popular AI coding assistants including Codex CLI, OpenClaw, Claude Code, and Oh My Pi agent. It directly addresses common frustrations with other projects—difficult installation, opaque inner workings, and inflated benchmarks—by offering a system that stores all user data locally in SQLite and markdown files. Its core promise is straightforward: it remembers what matters, forgets what doesn't, and intelligently surfaces relevant context over time.

Under the hood, Signet combines structured graphs, vector search, lossless compaction, and predictive injection. It is optimized for complete on-device operation, utilizing the nomic-embed-text model for embeddings and the lightweight 4-billion-parameter nemotron-3-nano model for background extraction and distillation. This local-first architecture ensures privacy and keeps the tool free and accessible, though it supports a bring-your-own-key (BYOK) option. Early performance is promising, with the system scoring 87.5% on a preliminary LoCoMo (Long Context Memory) evaluation. Signet is available now as open-source software for Windows, macOS, and Linux, with larger benchmark runs currently in progress.