The latest articles on DEV Community by DeRucci69 (@derucci69). https://dev.to/derucci69 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3293504%2Fe7066936-314b-49e2-80ef-e906c1d97fde.jpg https://dev.to/derucci69 en DeRucci69 Wed, 25 Jun 2025 10:42:16 +0000 https://dev.to/derucci69/runtime-api-discovery-for-llm-agents-using-a-simple-agentsjson-4dno https://dev.to/derucci69/runtime-api-discovery-for-llm-agents-using-a-simple-agentsjson-4dno <p>Current LLM tool use assumes compile-time bindings — every tool must be known in advance, added to the prompt, and hardcoded in.</p> <p>We built Invoke, a lightweight framework that lets agents discover and invoke APIs dynamically at runtime using a simple agents.json descriptor — no plugins, no schemas, no registries.</p> <p>The LLM uses a single universal function and discovers available tools just like a browser loading links.</p> <p>whitepaper: <a href="https://blog.invoke.network/you-dont-know-what-you-want-until-you-want-it/" rel="noopener noreferrer">https://blog.invoke.network/you-dont-know-what-you-want-until-you-want-it/</a></p> <p>Github: <a href="https://github.com/mercury0100/invoke" rel="noopener noreferrer">https://github.com/mercury0100/invoke</a></p> <p>1-minute demo: <a href="https://www.youtube.com/watch?v=CQISrRpyigs" rel="noopener noreferrer">https://www.youtube.com/watch?v=CQISrRpyigs</a></p> <p>Would love feedback and ideas — especially if you’re working on LLM agents or LangChain-style tooling.</p> api ai llm opensource