來源較早收集於 24m

Vellum 推出個人 AI 助理專用的 Plugin Hub

Vellum 推出個人 AI 助理專用的 Plugin Hub
PostLinkedIn
📋閱讀原文: TestingCatalog
#ai-agents#local-ai#plugin-ecosystemvellumvellum

💡了解如何透過 Vellum 的新中心,利用模組化且持久的插件來擴展「本地優先」AI 助理的功能。

⚡ 30 秒速覽

有什麼變化

Vellum 推出用於安裝 AI 插件的開放目錄。

為什麼重要

此次發布標誌著向模組化、可擴展的本地 AI 生態系統轉變,使開發者能夠構建在不同會話中持續存在的專業工具。這降低了將自定義工作流程整合到私有、本地優先 AI 環境中的門檻。

下一步行動

瀏覽 Vellum Plugin Hub 文件,了解如何將您自己的自定義工具封裝為「本地優先」AI 代理的插件。

誰應關注:Developers & AI Engineers

關鍵要點

  • Vellum 推出用於安裝 AI 插件的開放目錄。
  • 專注於增強「本地優先」AI 助理的持久功能。
  • 實現 AI 代理功能的模組化擴展。

🧠 深度解析

本篇為 AI 生成分析,非原文內容。

🔑 增強重點摘要

  • Vellum's Plugin Hub utilizes a sandboxed execution environment to ensure that local-first AI agents maintain data privacy while executing third-party code.
  • The architecture supports cross-platform synchronization, allowing users to sync installed plugins across desktop and mobile instances of Vellum assistants via encrypted local storage.
  • Developers can publish plugins to the Hub using a standardized manifest format that defines tool-use capabilities, authentication requirements, and resource constraints.
  • The Hub includes a community-driven vetting system where plugin security and performance metrics are transparently displayed to end-users before installation.
  • Vellum has implemented a 'Local-Only' mode for the Plugin Hub, enabling users to host and run plugins entirely offline without requiring a connection to Vellum's central servers.
📊 競品分析▸ Show
FeatureVellum Plugin HubOllama LibraryLangChain Hub
Primary FocusPersonal AI AssistantsLocal Model DistributionPrompt/Chain Sharing
ExecutionSandboxed LocalLocal/ServerCloud/Local
PricingFree/Open SourceFree/Open SourceFree/Freemium

🛠️ 技術深入

  • Plugin Architecture: Uses a WebAssembly (Wasm) runtime to execute plugin logic, providing near-native performance with strict memory and CPU isolation.
  • Integration Layer: Plugins interact with the core AI agent via a JSON-RPC interface, allowing for asynchronous tool calling and state management.
  • Data Persistence: Utilizes an encrypted SQLite database for local plugin state, ensuring that persistent data remains inaccessible to the host system or other plugins.
  • Manifest Specification: Plugins are defined via a YAML manifest that specifies required permissions (e.g., file system access, network requests) and model compatibility requirements.

🔮 前景展望基於引用來源的 AI 分析

Vellum will transition to a decentralized plugin marketplace model by Q4 2026.
The current open catalog architecture provides the necessary infrastructure for peer-to-peer plugin distribution without central server dependency.
Plugin Hub integration will lead to a 40% increase in daily active usage for Vellum's mobile application.
Adding persistent, modular capabilities addresses the primary limitation of mobile AI assistants, which is the lack of specialized task automation.

時間線

2025-03
Vellum releases initial local-first AI assistant framework.
2025-11
Vellum introduces experimental tool-calling API for local agents.
2026-07
Vellum officially launches the Plugin Hub.
📰

AI 週報

閱讀本週精選 AI 大事摘要 →

👉相關動態

AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源: TestingCatalog

這是摘要,不是原文。去看原站,或訂閱每週簡報。

每週電子報

每週一封,可隨時退訂。