Claude‑Mem is a persistent memory system for Claude Code that automatically records tool usage observations, creates semantic summaries, and makes them available in later sessions so the model retains project knowledge across restarts. Installation is performed with a single command such as npx claude‑mem install --ide grok‑bot (or variants for opencode, antigravity‑cli, OMP) which sets up a local worker service, SQLite storage, and optional real‑time feeds to Telegram, Discord, or Slack. After signing in via a magic‑link email, users receive a free 14‑day trial that can deliver up to 100 % more token usage before falling back to their Anthropic plan unless they subscribe; the trial can be bypassed with explicit provider flags or CI mode. The architecture relies on six lifecycle hook scripts (SessionStart, UserPromptSubmit, PostToolUse, Stop, SessionEnd) and a Bun‑managed worker exposing an HTTP API and web viewer. Memories are stored in SQLite and indexed with a hybrid Chroma vector database for semantic plus keyword search. The mem-search skill follows a three‑layer workflow: search returns compact IDs (~50‑100 tokens/result), timeline provides chronological context, and get_observations fetches full details for selected IDs (~500‑1,000 tokens/result), yielding roughly tenfold token savings. Configuration resides in ~/.claude‑mem/settings.json , supporting environment variables such as CLAUDE_MEM_MODE (e.g., code--zh for Simplified Chinese) to control workflow behavior and language. Required dependencies include Node.js 20+, the latest Claude Code with plugin support, Bun (auto‑installed), and uv for Python‑based vector search; SQLite 3 is bundled. The project is released under Apache 2.0 and maintains stable, core‑dev, and community‑edge branches.