ClawVault

**Judgement:** ClawVault appears to be a mature, well-engineered memory system for OpenClaw agents, suitable for serious and even production-like use by technically comfortable users who want durable, structured agent memory and are willing to manage a CLI- and hook-based workflow. **What it does well** - Provides a comprehensive memory layer: structured markdown vault, typed memories (decisions, lessons, relationships, projects, etc.), inbox capture, handoffs, wake/sleep/checkpoint/recover, and graph-aware context retrieval. - Integrates tightly with OpenClaw via hooks and dedicated diagnostics (`clawvault compat`, `doctor`), plus a well-defined folder structure and context profiles tuned to common agent situations (planning, incident, handoff). - Uses a local-first, auditable design: all memory is on the local filesystem; the skill bundle includes SKILL.md, HOOK.md, and the hook handler so users can inspect behavior before enabling. - Offers operational tooling for robustness, including a `repair-session` command to fix known transcript pathologies and automatic backups/dry-run modes. **Key risks / caveats** - **High-privilege filesystem access:** By design, it reads/writes extensively in the vault directory and modifies OpenClaw session transcripts. Misconfiguration or bugs could corrupt data, though backups and dry-runs mitigate this. - **External LLM dependency for `observe`:** The optional observation-compression flow calls Gemini (via GEMINI_API_KEY). This can leak transcript content to a third-party LLM; it must be disabled/avoided for strictly air-gapped or highly regulated environments. - **Operational complexity:** Setup requires npm, qmd, OpenClaw CLI, hook installation/enabling, and environment configuration. It’s best suited to users comfortable with CLI tools, inspecting JS source, and managing their agent runtime. **Recommended scenarios** - Teams or power users running OpenClaw agents that need long-lived, inspectable memory and stable cross-session continuity. - Workflows where agents must track projects, people, decisions, and lessons over time with clear handoffs and recovery from context loss. - Environments where local, markdown-based storage is preferred and users are willing to review hooks and manage security posture around the optional Gemini integration. **Less ideal for** - Users seeking a minimal, zero-setup memory layer or those uncomfortable enabling filesystem-writing hooks. - Strictly regulated or air-gapped deployments unless the `observe`/Gemini integration is disabled and the hook’s behavior is carefully audited.