OpenMem - Longterm Compressed Memory

SQLite long-term memory compression system for extended memory life. Adds tools for agents to control their memory functions.

This skill is internally coherent and implements the memory-compression features it describes, but it performs privacy-sensitive and destructive actions. Before installing: 1) Back up your OpenClaw session directory and any existing memory DB (the installer and uninstall script do not delete the DB). 2) Review and, if needed, override OPENMEM_SESSIONS_DIR and OPENMEM_DB to point to safe locations. 3) If you do not want automatic deletion of raw transcripts, do not enable auto-compress or run auto_compress.py with --no-wipe (or disable the cron). 4) Remember the cache file (openmem-cache.json) is plaintext and stored alongside the DB — treat it as sensitive. 5) The skill registers an MCP server and a cron job with your gateway; ensure you trust the skill before enabling hooks/cron. 6) Note minor inconsistencies in the metadata (the package contains runnable scripts despite saying “instruction-only” and SKILL.md references "Ollama" in places though the code invokes the openclaw CLI); if anything about provider usage or the model pipeline is unclear, inspect the local scripts (auto_compress.py, mcp_server.py, compress.py) before enabling automatic behavior.

Type: OpenClaw Skill Name: openmem Version: 1.2.5 OpenMem is a long-term memory management system that uses SQLite to store and search agent memories. The skill includes an automated compression feature (auto_compress.py) that reads session logs, extracts key information using the user's configured LLM, and replaces old session files with stubs to save space. While the session-wiping behavior is destructive, it is clearly documented as a core feature, includes a '--no-wipe' safety flag, and performs database verification before deletion. The code follows standard OpenClaw patterns for MCP servers and hooks, and no evidence of data exfiltration or unauthorized network activity was found.