Vault Operator是一款开源AI Agent产品，能够集成到用户知识库中，提供多层次的智能协作和隐私控制功能 [1]。该产品支持在编辑器中进行内联AI聊天 [1]，并可将网页内容剪辑转换为永久笔记 [1]，用户可选择使用本地或云端模型 [1]。
产品通过三层跨会话记忆机制提升任务连贯性：Soul层存储长期偏好，Facts层保存结构化陈述，History层记录对话历史 [1]。Vault Operator支持通过MCP服务器与Claude Desktop、ChatGPT、Perplexity等多种AI工具连接 [1]，并提供细粒度的审批和隐私控制——用户可配置自动审批规则，每次文件更改前系统显示差异对比 [1]。在实际应用中，该产品在简单搜索摘要任务上实现了显著的效率提升，token使用量从约634K降低至约60K [1]。作为开源产品，Vault Operator本身免费提供，用户仅需支付所使用AI模型的费用 [1]。
Vault Operator is an AI agent product designed to integrate directly into users' knowledge bases, offering a practical window into how contemporary AI systems function in practice [1]. The platform combines inline AI chat capabilities, semantic search, document generation, and persistent memory across conversations, all controlled through a dashboard that grants users fine-grained approval and privacy controls [1]. Users can choose to run the agent locally or in the cloud, and it learns from workflow patterns to improve task execution efficiency [1].
The agent architecture centers on three layers of cross-session memory to maintain context over time [1]. The "Soul" layer preserves long-term user preferences, the "Facts" layer stores structured statements, and the "History" layer retains conversation records [1]. Vault Operator connects to other AI tools via MCP servers, enabling integration with Claude Desktop, ChatGPT, and Perplexity [1]. The platform's inline chat functionality became available starting with version 3.0.0, while web clipping capabilities were added in version 3.3.6 [1]. A significant efficiency gain is evident in routine tasks: token consumption for simple search-and-summarize operations dropped from approximately 634,000 to roughly 60,000 [1]. Users can configure automatic approval rules and review file changes through diffs before implementation [1]. The agent itself is open source and free; users pay only for the AI models they consume [1].