SqueakGPT

January 1, 2025 ยท View on GitHub

A custom GPT for ChatGPT that talks to your Squeak image.

Setup

Server

  • Get a recent Squeak image and check out the SqueakWorkspaceServer package using the Git Browser.

  • Start the server:

    server := SqueakWorkspaceServer new.
server startOn: 8080.

"server stop."

  • If you do not want to expose your own port, you can use any kind of tunnel. For example, I used cloudflared:

    cloudflared tunnel --url http://localhost:8080

    If you use a tunnel, you must specify the public tunnel URL to the server:

    server tunnelServerUrl: 'https://your-unique-tunnel-name.trycloudflare.com'.

Warning


Remember that everybody who has access to your server can execute arbitrary code on your machine! Consider using strong authentication or keep the URL private.

ChatGPT

  • Create a new custom GPT and configure it based on gpt.yaml.
  • Replace <YOUR_SERVER_URL> by your actual server (or tunnel) URL.
  • Save the GPT (but beware of making it public or sharing it with others because it exposes remote access to your local machine!) and enjoy!

Impressions

You can read an example conversation here: https://chatgpt.com/share/67381691-7330-8004-b515-bf5c98416cd7

image image image

Development

ChatGPT

Chat mode of GPT Builder is very tedious. It is helpful for bootstrapping the instructions but often did a bad job for me at maintaining prior instructions. Be sure to review all changes with version control. Also, the instruction text is currently limited to 8000 characters. In the future, we should extract many best practices into a queryable knowledge base/memory.

See Also

If you want to use ChatGPT through the OpenAI API in Squeak instead, also check out SemanticText.