Building an MVP Agentic Tool-Use Bot with Node.js and OpenRouter 🌤️🤖

In the world of AI, building a ChatGPT-like chatbot capable of performing real-world tasks can be quite a challenge. However, the Mausam AI bot demonstrates how to create such an interface using Node.js, OpenRouter, and a tool-based approach called Agentic Tool-Use. This innovative architecture allows the LLM (Large Language Model) to fetch real-time weather data, such as temperature, weather, and humidity, from an external API (wttr.in) and utilize that information in its responses.

At the core of Mausam AI's functionality lies a system prompt and agent logic implemented in the helper.js file. This structure enforces a specific workflow: START ➡️ PLAN ➡️ TOOL ➡️ OUTPUT. By using a JSON output format, the bot's backend can process the AI's responses step by step, ensuring accurate and timely generation of information.

When the AI needs to utilize the get_mausam tool, it outputs a JSON object specifying the tool name and input location. The backend then executes the tool and fetches real-time weather data from wttr.in. This live data is injected back into the conversation context, enabling the LLM to generate precise, up-to-the-minute responses.

One key aspect of the bot's design is the MAX_ITERATIONS check, which prevents infinite loops or hallucinations during the agent's workflow. In addition, the bot includes a robust API call mechanism with a fallback system in place, ensuring reliable and uninterrupted service even if there are issues with the primary API key.