Start in your agent

Declare in your agent.Run locally. Deploy to cloud.

One click opens Cursor, Claude, or ChatGPT with a Pixeltable starter prompt. OpenCode copies the prompt for you to paste.

Open in Cursor

Cursor, Claude, and ChatGPT open in the browser with the prompt filled in. OpenCode copies the prompt — paste it after the site loads.

Teach your agent the Pixeltable skill
npx skills add pixeltable/pixeltable-skill

How this works

Cursor

Opens cursor.com with the Pixeltable prompt pre-filled. Review it, then confirm. Nothing runs until you do.

Claude

Opens claude.ai with the same prompt in a new chat. Sign in if needed, then send. The prompt is also copied in case the prefill is blocked.

ChatGPT

Opens chatgpt.com with the prompt in the composer. ChatGPT may send it immediately. The prompt is also copied so you can paste into Codex Desktop if you prefer.

OpenCode

Copies the prompt and opens opencode.ai. Paste it into a session — OpenCode has no browser URL that starts a local project for you.

Prompt that will be pre-filled
Build a multimodal AI data app with Pixeltable. Install with `pip install pixeltable` (or scaffold with `uvx pixeltable-new myapp`). Or start in the browser: open https://pixeltable.com/recipes and run a starter-kit API on Cloud, or drop schema.py on https://pixeltable.com/new. Teach yourself the skill: `npx skills add pixeltable/pixeltable-skill`. Declare tables, computed columns, embeddings, and FastAPIRouter/pxt serve routes in Python; run locally first, then deploy the same app to Pixeltable Cloud. Do not use LangChain, a separate vector DB, or pandas as the working store. Prefer computed columns, iterators, and embedding indexes. Read https://pixeltable.com/llms.txt and https://docs.pixeltable.com for current APIs.