This workflow dynamically chooses between two new powerful Anthropic models — Claude Opus 4 and Claude Sonnet 4 — to handle user queries, based on their complexity and nature, maintaining scalability and context awareness with Anthropic web search function and Think tool. --- Key Advantages 1. 🔁 Dynamic Model Selection Automatically routes each user query to either Claude Sonnet 4 (for routine tasks) or Claude Opus 4 (for complex reasoning), ensuring optimal performance and cost-efficiency. 2. 🧠 AI Agent with Tool Use The AI agent can utilize a web search tool to retrieve up-to-date information and a Think tool for complex reasoning processes — improving response quality. 3. 📎 Memory Integration Uses session-based memory to maintain conversational context, making interactions more coherent and human-like. 4. 🧮 Built-in Calculation Tool Handles numeric queries using an integrated calculator tool, reducing the need for external processing. 5. 📤 Structured Output Parser Ensures outputs are always well-structured and formatted in JSON, which improves consistency and downstream integrations. 6. 🌐 Web Search Capability Supports real-time information retrieval for current events, statistics, or details not available in the AI’s base knowledge. --- Components Overview Trigger: Listens for new chat messages. Routing Agent: Analyzes the message and returns the best model to use. AI Agent: Handles the conversation, decides when to use tools. Tools: websearch for internet queries Think for reasoning Calculator for math tasks Models Used: claude-sonnet-4-20250514: Optimized for general and business logic tasks. claude-opus-4-20250514: Best for deep, strategic, and analytical queries. --- How It Works 1. Dynamic Model Selection - The workflow begins when a chat message is received. The Anthropic Routing Agent analyzes the user's query to determine the most suitable model (either Claude Sonnet 4 or Claude Opus 4) based on the query's complexity and requirements. - The routing agent uses predefined criteria to decide: - Claude Sonnet 4: Best for standard tasks like real-time workflow routing, data validation, and routine business logic. - Claude Opus 4: Reserved for complex scenarios requiring deep reasoning, advanced analysis, or high-impact decisions. 2. Query Processing and Response Generation - The selected model processes the query, leveraging tools like websearch for real-time information retrieval, Think for internal reasoning, and Calculator for numerical tasks. - The AI Agent coordinates these tools, ensuring the response is accurate and context-aware. A Simple Memory node retains session context for coherent multi-turn conversations. - The final response is formatted and returned to the user without intermediate steps or metadata. --- Set Up Steps 1. Node Configuration - Trigger: Configure the "When chat message received" node to handle incoming user queries. - Routing Agent: Set up the "Anthropic Routing Agent" with the system message defining model selection logic. Ensure it outputs a JSON object with prompt and model fields. - AI Model Nodes: Link the "Sonnet 4 or Opus 4" node to dynamically use the selected model. The "Sonnet 3.7" node powers the routing agent itself. 2. Tool Integration - Attach the "websearch" HTTP tool to enable internet searches, ensuring the API endpoint and headers (e.g., anthropic-version) are correctly configured. - Connect auxiliary tools (Think, Calculator) to the "AI Agent" for extended functionality. - Add the "Simple Memory" node to maintain conversation history. 3. Credentials - Provide an Anthropic API key to all nodes requiring authentication (e.g., model nodes, web search). 4. Testing - Activate the workflow and test with sample queries to verify: - Correct model selection (e.g., Sonnet for simple queries, Opus for complex ones). - Proper tool usage (e.g., web searches trigger when needed). - Memory retention across chat turns. 5. Deployment - Once validated, set the workflow to active for live interactions. ---- Need help customizing? Contact me for consulting and support or add me on Linkedin.
