neboy72/openclaw-nexus-memory

OpenClaw Nexus Memory is a Python toolkit for taking care of the long-term memory of an AI agent. The agents it targets, called OpenClaw agents, store their memory as plain Markdown files such as MEMORY.md plus a memory and wiki folder. Over time those notes go out of date or start to contradict each other. Nexus runs health checks against those files, can fix some problems automatically, and adds a wiki layer on top, all without needing a database. There are three main pieces. A health script scans the Markdown for stale patterns (facts that are out of date, defined by regex rules in a YAML config), contradictions (pairs of patterns that conflict, such as one file saying a feature is enabled and another saying it is disabled), and gives an overall health score from 0 to 10. A patterns script looks for knowledge gaps, which are entities mentioned a lot but never given their own wiki page, finds co-occurrences between entities, and tracks how the memory grows over time. Auto-fix is the action half of the tool. When you pass --auto-fix, stale MEMORY.md entries are marked as HISTORICAL, RESOLVED, or ARCHIVED, which excludes them from later scans so old incidents stop being flagged as current problems. A --dry-run flag previews the changes before anything is written. There is also a scheduled maintenance recipe using OpenClaw's cron feature so the weekly run happens on its own and delivers a report through a channel like Telegram. Installation is either a curl install script or a self-install where you point the agent at an AGENTS.md URL and it does the setup itself. For semantic work, embeddings can come from sentence-transformers by default (local, no key), Ollama as a local service, or the Voyage cloud API. The wiki layer ships Obsidian-compatible templates for entities like Agent, User, Infrastructure, and Concepts, linked with Wikilinks. Requirements are Python 3.11 or newer, and the README is clear that no Qdrant or other vector database is needed.