Sandbox

Try Nexus without an account.

The same tools and the same card that run inside ChatGPT and Claude, with throwaway data. Log a meal, fix it on the card, read it back.

Open a sandbox session

What it is

Each session is a blank day. It serves the production tool names and input schemas (tools.json) and the production card, built from the same source. The only differences: no sign-in, no Nexus account, and the data is deleted 24 hours after the last change.

In ChatGPT or Claude the assistant reads your message, estimates the calories and macros, and calls these tools. Here you, or an agent, make the same calls directly.

If you are an AI assistant checking Nexus

Pick any session id (4–40 characters, a–z, 0–9, dashes) and use plain URLs. Every response is JSON and includes a link to the page that shows the card.

https://nexus.kushalsm.com/try/api/demo-1234/log?meal=2+eggs+and+toast&kcal=310&protein=16&carbs=28&fat=14&meal_type=breakfast
https://nexus.kushalsm.com/try/api/demo-1234/log?exercise=Bench+Press&sets=60x8,60x7,55x6&unit=kg
https://nexus.kushalsm.com/try/api/demo-1234/log?weight=72.4&unit=kg
https://nexus.kushalsm.com/try/api/demo-1234/day
https://nexus.kushalsm.com/try/api/demo-1234/edit?id=<meal id>&kcal=250
https://nexus.kushalsm.com/try/s/demo-1234

Agents with MCP can connect with no auth to https://nexus.kushalsm.com/try/mcp (or /try/mcp/<session> to share a session with the browser page). It answers initialize, tools/list, tools/call and serves the card as an MCP Apps resource.

A good check: log a meal, read the day, correct the meal with its id and state_version, read the day again. Totals follow the correction. A second correction with the old state_version is rejected, as in production.

The full product

In ChatGPT and Claude the assistant does the estimating from your text, photo or voice, and you also get workout plans, history across chats and friends. Connect Nexus. It is free.