mattvr / shellgpt Goto Github PK

Even after changing the model to gpt-3.5-turbo using gpt --config. I am using MacOS 13.3.1

I keep getting this error:

Failed to parse message TypeError: Failed to parse message:
"{
    "error": {
        "message": "The model: `gpt-4` does not exist",
        "type": "invalid_request_error",
        "param": null,
        "code": "model_not_found"
    }
}
"

Could the key be provided in the config as an alternative? Also, should the check only occur when making a request? As it stands, if the key isn't found, all commands produce an error, including the 'help' and configuration commands.

I like this deno-based version and find it easy to install and interact with, but I wonder if further features could make it even more powerful.
For example, it would be great if ShellGPT was OS and shell aware like this tool for example: https://github.com/TheR1D/shell_gpt
I also feel there could also other aspects of https://github.com/TheR1D/shell_gpt that could be used for this deny-based ShellGPT tool.

This project rocks !!!

$ deno install -frA -n gpt https://deno.land/x/shellgpt/mod.ts
Would you like to run the above command to complete ShellGPT installation?
(y)es [default]
(n)o
leave blank to use default

y

error: Uncaught (in promise) NotFound: program not found
const command = Deno.run({
^
at opRun (ext:runtime/40_process.js:50:14)
at Object.run (ext:runtime/40_process.js:136:15)
at install (https://deno.land/x/[email protected]/lib/update.ts:146:24)
at https://deno.land/x/[email protected]/install.ts:61:22
at eventLoopTick (ext:core/01_core.js:183:11)

First, I love ShellGPT. I use it a ton. Thank you!

Second, I wish there were a way in the REPL to paste in content, or type in content, which has line breaks. Is there a way to have line breaks in input in the REPL? Perhaps a different key combo (Ctrl+Enter?) could be used to submit a request?

I think it would require replacing prompt with stdin/stdout handling. I've never done much with CLI apps in Deno but if I'm feeling adventurous and have the time I might give it a try myself.

Requested by a few people. Any more details or suggestions of specific models to use would be helpful before integrating.

Deno.run has been deprecated and replaced with Deno.command. See https://deno.land/[email protected]?s=Deno.Command

I did try fixing it in a fork, but when I checked the current code out it would give me this error message even before I changed any code:

(No message passed)

{
  "error": {
    "message": "[] is too short - 'messages'",
    "type": "invalid_request_error",
    "param": null,
    "code": null
  }
}

Failed to parse message TypeError: Cannot read properties of undefined (reading '0')
    at Object.next (file:///C:/Users/paul/Downloads/shellgpt/ShellGPT/lib/ai.ts:150:31)
    at eventLoopTick (ext:core/01_core.js:183:11)
    at async file:///C:/Users/paul/Downloads/shellgpt/ShellGPT/mod.ts:537:34

When I set up the program and export my openAI key via the install instructions, I get the following message when trying to use the program:

Failed to parse message TypeError: Failed to parse message:
"{
    "error": {
        "message": "The model: `gpt-4` does not exist",
        "type": "invalid_request_error",
        "param": null,
        "code": "model_not_found"
    }
}
"

TypeError: Cannot read properties of undefined (reading '0')
    at Object.next (https://deno.land/x/[email protected]/lib/ai.ts:179:31)
    at async https://deno.land/x/[email protected]/mod.ts:311:30`

I see the other closed issue and used the install line from there via deno, but issue persists.
Using MacOS and installed deno via brew

I'm realizing w/ gpt refactoring can happen at a whole new level now? I guess we call it semantic refactoring? The only problem is token limits?

Are there things we could do via an lsp client (language server protocol) which would help w/ token limitations? While allowing larger scale refactoring? ie lsp-mode in emacs might be a good starting point?

Would it make sense to enable discussions on this repo? For dumb questions like this? ;)

Given gpt cannot yet handle unified diffs, perhaps simple search/replace strings could suffice in the meantime? For automated editing/patching of existing codebases?

openai/evals#537

Again, this project rocks !!! Love it!

It's a little unclear to me why you'd want to throttle the output to a given wpm. Is there a way to configure this limit to be "off"?

Please add an option to API into local OpenAI compatible API server (llama.cpp.python[server] for example)