← Back to Catalog

Versatly/clawvault

↗ GitHub

🐘 An elephant never forgets. Structured memory system for AI agents.

634

Stars

60

Forks

4

Watchers

8

Open Issues

TypeScript·MIT License·Last commit Mar 26, 2026·by @Versatly·Published April 1, 2026·Analyzed 6d ago
B

Safety Rating B

The repository appears to be a legitimate open source AI agent memory system. However, the README contains a reference to a meme coin token ('$CLAW') on pump.fun, which is an unusual inclusion in a developer tooling project and may indicate financial promotion or a speculative token scheme associated with the project. This does not constitute malicious code, but users should be aware of this association. No hardcoded secrets, malicious code patterns, or prompt injection attempts were identified. Dependency manifests were not available for review.

AI-assisted review, not a professional security audit.

AI Analysis

ClawVault is a local-first, markdown-native structured memory system for AI agents. It provides persistent context across sessions using a file-based vault of categorized markdown documents, a typed knowledge graph built from wiki-links, hybrid BM25 + semantic search, and a set of 8 core primitives (Goals, Agents, State Space, Feedback, Capital, Institution, Synthesis, Recursion) modeled as CLI commands. It integrates with Obsidian for visualization, supports multiple LLM providers for compression and context generation, and offers session lifecycle management (wake/checkpoint/sleep) to prevent 'context death' between agent sessions.

Use Cases

  • Providing AI agents with persistent memory that survives session resets and context window limits
  • Building a local, human-readable knowledge base of decisions, lessons, and project state for AI-assisted workflows
  • Integrating structured agent memory into Obsidian for graph visualization and Kanban task management
  • Enabling multi-session continuity for autonomous agents via checkpoint/recover primitives
  • Performing hybrid semantic + keyword search over accumulated agent memories and notes

Tags

#memory#ai-agents#knowledge-graph#rag#local-first#cli-tool#llm#embeddings#workflow-automation#plugin

Security Findings (4)

hardcoded_secrets

The README contains a reference to what appears to be a cryptocurrency token contract address ('$CLAW': 5Fjr82MTB8mvxkzi9FYtvrUsPiDGE2M29w3dYcZpump on pump.fun), suggesting a meme coin associated with the project. This is not a hardcoded secret but is a notable finding regarding potential financial promotion embedded in an open source project.

malicious_code

No malicious code patterns are visible from the repository content provided. All described functionality relates to memory storage, search, and session lifecycle management.

dependency_vulnerabilities

No manifest files (package.json, etc.) were provided for analysis; dependency vulnerability assessment could not be performed.

prompt_injection_attempt

No prompt injection attempts detected in the README or provided content.

Project Connections