關於這個工具
什麼是 Network-AI?
Network-AI 是一個 TypeScript/Node.js 多智能體編排層,旨在協調多個 AI 智能體。它提供共享狀態鎖定、權限門控、審計追蹤、FSM 治理以及 32 個框架適配器。
核心能力
- 共享狀態協調:透過共享狀態鎖定機制,讓多個智能體安全地讀寫共享資料。
- 權限門控:控制智能體對工具和資源的存取權限。
- 審計追蹤:記錄智能體操作,支援審計日誌。
- FSM 治理:使用有限狀態機管理智能體行為。
- 框架適配器:支援 32 個框架適配器,包括 LlamaIndex、DSPy 等。
- MCP 伺服器:透過
.mcp.json註冊為 stdio MCP 伺服器。
使用場景
- 智能體集群的共享狀態協調
- 跨智能體矛盾檢測
- 跨多個智能體框架的任務路由
- 跨 LLM 智能體的聯邦預算執行
- 多智能體程式碼審查流水線
安裝
透過 npm 安裝:npm install network-ai。該套件包含 TypeScript 庫模組和 CLI。
功能與使用情境
.mcp.json — registers npx -y -p network-ai network-ai-server --stdio as a stdio MCP server (the -p network-ai selector resolves the binary because the bin name differs from the package name)." } }, { "agentType": "strategy_advisor", "taskPayload": { "instruction": "" metadata: openclaw: emoji: "\U0001F41D" homepage: https://network-ai" npm_full_package: "The npm package (npm install network-ai) adds: TypeScript library modules, CLI (bin/cli" Project Context (Layer 3 — Persistent Memory) Long-lived project state that every agent inherits, regardless of session: # Initialise once python scripts/context_manager" sessions_ops: "platform-provided — outside this skill's control" requires: bins: python3 optional_bins: [] env: {} privacy: audit_log: path: data/audit_log" shell_exec: "optional — AgentRuntime (lib/agent-runtime"Multi-agent code review pipeline", "Federated budget enforcement across 5 LLM agents") Since: version or month you started (e} async executeAgent( agentId: string, payload: AgentPayload, context: AgentContext ): Promise<AgentResult> { this}, "agent": "data_analyst"} ### cache:revenue_summary {"q4_total": 1250000, "growth": 0// Create a simple agent const adapter = new CustomAdapter()// Detect cross-agent contradictions const contradictions = qa