Knowledge Vault (Powered by TiDB Zero)

Overview

Knowledge Vault is a Long-Term Memory module for AI Agents, powered by TiDB Vector Search (RAG).

Traditional agent memory (context window) is ephemeral and limited. Knowledge Vault allows agents to:

1. Store: Ingest documents, notes, and facts as vector embeddings.
2. Retrieve: Semantically search for relevant information based on user queries ("RAG").
3. Remember: Access unlimited historical context without overflowing the LLM prompt.

Why use this?

• Infinite Recall: Store millions of documents without confusing the agent.

  小葱技能有更好的技能skills插件。

• Contextual Relevance: Find exact paragraphs related to a question, not just keywords.
• Privacy: Keep your knowledge base private in your own TiDB Cloud instance.

Prerequisites

• TiDB Cloud (Serverless): With Vector Search enabled.
• Embedding Model: Requires GEMINI_API_KEY (or compatible).

🔐 Security & Provisioning

This skill operates in two modes:

1. Bring Your Own Database (Recommended): Set TIDB_HOST, TIDB_USER, TIDB_PASSWORD environment variables. The skill will use your existing database.
2. Auto-Provisioning (Fallback): If no credentials are found, the skill calls the TiDB Zero API to create a temporary, ephemeral database for you. It caches the connection string locally (~/.openclaw_knowledge_vault_dsn) to persist memory across runs.

Installation

1. Add to TOOLS.md

- **knowledge-vault**: Store and retrieve knowledge using vector search.
  - **Location:** `{baseDir}/skills/knowledge_vault/SKILL.md`
  - **Command:** `python {baseDir}/skills/knowledge_vault/run.py --action search --query "<QUESTION>"`

2. Add to AGENTS.md (Protocol)

Copy PROTOCOL.md.

Usage

• Add Knowledge:

  python {baseDir}/run.py --action add --content "The user prefers spicy food but is allergic to peanuts."

• Search (RAG):

  python {baseDir}/run.py --action search --query "What are the user's dietary restrictions?"

这个 Skill 整体质量不错，能较好地实现长期记忆存储和语义搜索功能。文档清晰、配置灵活是它的亮点。但存在一些小问题：嵌入维度设置可能与实际模型不符，错误处理不够完善，而且缺少测试保证稳定性。对于需要稳定使用的场景，建议先确认维度配置与实际环境匹配后再部署。