ToolOps: 프로덕션용 AI 에이전트에서 하나의 데코레이터 사용

AI Agents are only as reliable as the tools they wield. In production, tools are expensive, unreliable, and slow. ToolOps is a framework-agnostic middleware SDK that treats every tool call as a first-class operation. By wrapping your tools in a single decorator, you instantly upgrade them with industrial-grade caching, resilience, and observability. "ToolOps is to AI Tools what Service Mesh is to Microservices." | Feature | Standard @lru_cache | ToolOps SDK | |---|---|---| | Async Support | ❌ No | ✅ Native | | Semantic Cache | ❌ No | ✅ Yes (Embeddings) | | Resilience | ❌ No | ✅ Circuit Breaker / Retries | | Distributed | ❌ No | ✅ Postgres (S3 / Redis coming) | | Observability | ❌ No | ✅ OTEL / Prometheus | | AI Native | ❌ No | ✅ MCP / LangChain / LangGraph / CrewAI etc. Ready | Every agent developer hits the same wall when moving from demo to production: | The Problem | The Business Impact | The ToolOps Fix | |---|---|---| | Redundant Calls | 💸 Skyrocketing API costs | Semantic Cache (1 call, 99 hits) | | API Instability | 💥 Agent crashes & loops | Circuit Breaker & Retries | | Concurrency | 🐢 Bottlenecks on shared tools | Request Coalescing | | Observability | 🌑 Blind operations | Native OTEL & JSON Logging | | Standardization | 🧩 Framework lock-in | Universal Decorator | Install the SDK (Core is zero-dependency): pip install toolops[all] Supercharge your tools in seconds: from toolops import readonly, sideeffect, cache_manager from toolops.cache import MemoryCache, PostgresCache # 1. Plug-and-play Backends cache_manager.register("fast", MemoryCache(), is_default=True) # 2. Add Intelligence & Resilience to any function @readonly(cache_backend="fast", cache_ttl=3600, retry_count=3) async def get_market_data(ticker: str): return await api.fetch(ticker) # Automatically cached & retried @sideeffect(circuit_breaker=True, timeout=5.0) async def execute_trade(order: dict): return await broker.submit(order) # Protected by Circuit Breaker Don't just cache exact strings. ToolOps uses vector embeddings to understand the meaning of a query. If an agent asks "How's the weather in Paris?" and then "What is the Parisian weather like?", ToolOps serves the cached result. Reduces LLM latency by up to 90%. - Circuit Breakers: Stop pounding failing APIs before they take down your system. - Stale-if-Error: If an upstream API fails, ToolOps can automatically serve the last known good value from the cache. - Request Coalescing: If 50 agents call the same tool simultaneously, ToolOps executes it once and multicasts the result. ToolOps is "AI-Native". It includes built-in support for MCP, allowing you to instantly expose your decorated tools to Claude Desktop, Cursor, or any MCP-compatible host without writing a single line of JSON Schema. ToolOps is designed to be the "glue" of the AI ecosystem. It works natively with: - LangChain / LangGraph - CrewAI - LlamaIndex - Model Context Protocol (MCP) - PydanticAI - AutoGPT - Any Python function-based agent framework ToolOps doesn't just run your tools; it measures them. - Structured Logging: Every hit, miss, failure, and retry is logged in production-ready JSON. - OpenTelemetry: Native traces and spans to visualize tool execution in Jaeger, Honeycomb, or Datadog. - Prometheus: Real-time metrics for cache hit rates and tool latency. - Web Dashboard: Real-time metrics, cost attribution, and hit rates UI. - Budget Control: Hard limits on tool-induced API costs per hour/day. - Native MCP Server: One-click deployment of ToolOps tools as a standalone host. - Streaming Middleware: Support for streaming tool outputs in real-time. - New Backends: MariaDB, ChromaDB, and Pinecone support. ToolOps includes a command-line tool to inspect and manage your tool infrastructure. # See all available commands toolops --help # Check system health and backend readiness toolops doctor # View real-time cache statistics toolops stats --app my_app:setup_toolops # Clear a specific cache backend toolops clear postgres --app my_app:setup_toolops ToolOps is an open-source project by Hedi MANAI. I am building the future of Agentic Operations through lightweight, industrial-grade tools. - LinkedIn Connect with me - GitHub Follow my work - Website Hedi Manai Portfolio - X (Twitter) Follow @hedi_manaii - Discord @hedimanai - Reddit Profile - Contact [email protected] Apache License 2.0 — see LICENSE for details. Empowering the next generation of Agentic Workflows.