ClawVault addresses AI agent "context death" by providing a local-first, structured memory system. It uses markdown as its storage primitive, offering human-readable, git-friendly, and graph-aware context retention between sessions, preventing data loss and enhancing agent understanding.

How It Works

This system uses a markdown-native, local-first architecture for data privacy and no vendor lock-in. It auto-generates a knowledge graph from wiki-links/frontmatter for richer context retrieval. Session resilience is achieved via checkpoint/recovery primitives. ClawVault is built around 8 core primitives (Goals, Agents, State Space, etc.) mapping to its CLI commands and vault structure for coherent memory management.

Quick Start & Requirements

Install via npm: npm install -g clawvault and npm install -g github:tobi/qmd. Requires Node.js 18+ and qmd on PATH. LLM API keys (Anthropic, OpenAI, etc.) needed for advanced features. Setup estimated at 5 minutes. Docs: clawvault.dev, npm: npmjs.com/package/clawvault.

Highlighted Details

• Markdown-Native & Local-First: Human-readable storage, data stays local.
• Knowledge Graph: Auto-generates graph from links/frontmatter for richer context.
• Session Resilience: Checkpoint/recovery primitives prevent data loss.
• Context Profiles: Tune retrieval for specific tasks (e.g., planning, incident).
• Obsidian Integration: Seamlessly works with Obsidian for graphs, tasks, Kanban.
• OpenClaw Integration: Hook-based lifecycle management for automated actions.
• Performance: 57% overall on LongMemEval, 71% for Session Continuity.

Maintenance & Community

Features 6 external contributors and 20+ merged PRs. Active development since February 2026. Details on GitHub.

Licensing & Compatibility

Released under the MIT license, permitting commercial use and closed-source linking.

Limitations & Caveats

Requires careful integration between MEMORY.md and the vault. Benchmark scores indicate competitive but not leading performance in all areas, particularly memory retention and context retrieval.