Build an AI Agent That Calls Live Tools | Free Sandbox
This template shows an AI agent as an ordinary workflow step, wired to two live external tools it can call for itself. A request arrives on a webhook carrying a question like "What's the weather in Seattle?", the agent reasons over it, and capability edges grant it two tools: a Geocoder (OpenStreetMap Nominatim) that turns a place name into coordinates, and a Weather Forecast tool (MET Norway) that returns the forecast for those coordinates. The agent-to-tool connections are grants, not control flow: the model decides which tool to call, with what arguments, and in what order — here it geocodes first, then forecasts, then answers in plain language. Both APIs are keyless, so the demo makes real outbound calls with no credentials. The demo request is pre-filled, so running it exercises the whole two-tool chain end to end. Everything runs in a throwaway sandbox workspace on the platform's own AI gateway, so you can try the agent-plus-tools pattern without an account or an AI key of your own.
How to test AI Agent with Tools
This sandbox opens with a real example payload already loaded into a live workflow. Run it to watch the event move through every node, inspect the JSON at each step, and edit the mapping logic in place — no sign-up, no setup, no local tunnel. Create a free account when you are ready to point your own webhooks at a live endpoint and keep the workflows you build here.
- Start the sandbox to load a real AI Agent with Tools example payload into a live workflow.
- Run the workflow and watch the event move through each node on the canvas.
- Edit the mapping, add a destination, and re-run to see your changes immediately.