EncrEor/rlm-claude

pip install mcp-rlm-server[all]
uv tool install mcp-rlm-server[all] --python 3.12
git clone && ./install.sh (full install with hooks)
docker build -t rlm-server . (Docker)

Usage

• •rlm_recall(importance='critical', limit=50) at session startup to load universal rules
• •rlm_remember('...', category='decision', importance='critical', tags='...') to save insights
• •rlm_chunk(...) to save conversation segments; rlm_peek / rlm_grep / rlm_search to retrieve
• •rlm_retention_preview / rlm_retention_run / rlm_restore for archive lifecycle
• •/rlm-analyze and /rlm-parallel sub-agent skills

Key features

• ✓Auto-save snapshot before /compact via PreCompact hook
• ✓14 tools: remember, recall, forget, status, chunk, peek, grep, search, list_chunks, sessions, domains, retention_preview, retention_run, restore
• ✓Hybrid BM25 + cosine semantic search with Model2Vec (default) or FastEmbed providers
• ✓3-zone retention lifecycle with immunity system (critical tags, frequent access, keywords)
• ✓Multi-project organization with auto-detection from git/working directory
• ✓Typed chunks (snapshot/session/debug), FR/EN/JA i18n
• ✓Security: path traversal prevention, atomic writes, file locking, SHA-256 dedup

Best for

Claude Code users who lose context after /compact and want persistent searchable memory

Caveats

• ⚠Requires Python 3.10+ and Claude Code CLI
• ⚠Hooks must be configured in ~/.claude/settings.json
• ⚠Semantic search requires optional extra installs