-
Notifications
You must be signed in to change notification settings - Fork 3.2k
Frequently Asked Questions
These are some of the most asked questions about GPT Pilot. If you don't find an answer here, try asking in our Discord community. Also check out our video tutorials.
While you can configure GPT Pilot to use other models, including local LLMs, it probably won't work well. This is because GPT Pilot prompts are optimized for GPT4, and can easily confuse other models (including GPT3.5). If you use other model and GPT Pilot seems to get stuck, this is the probable reason.
If you're willing to experiment, you can modify the LLM prompts in your GPT Pilot installation and see how far you can get with other models. Several people in our Discord community are interested in this so if you're feeling adventurous, join our Discord to learn and share your hacks.
The VSCode extension asks me to log in? What am I logging into? Do I need to use my OpenAI email and password?
This is separate from your OpenAI email/password - don't reuse them (you can use the same email address, but the accounts are separate, and you should never reuse passwords anyways). We're building a cloud version of GPT Pilot so the VSCode extension already signs up to our system. We also use this to ask you for feedback (how was GPT Pilot for you).
I entered my OpenAI key but I get "you don't have access to gpt-4" or "openai model not found" error
Please note that ChatGPT Plus subscription and OpenAI API subscriptions are separate. To use GPT Pilot with GPT4, you need to have a paid subscription to OpenAI API. To verify this, you can go to OpenAI API Playground and check if you can use gpt-4-1106-preview model.
If you can use gpt4 from the playground, double-check that the API key you pasted is correct.
When I start GPT Pilot, it asks me some questions and then offers advice how to start coding, but doesn't actually do anything
If you're using an AI model other than GPT4, you've run into our compatibility issue (see Do you support AI models other than OpenAI GPT4? question).
Check that you're running Python 3.9 or later. If you have multiple Python installations on your system, GPT Pilot might have picked up the incorrect version. You can fix this by manually setting up Python virtual environment in gpt-pilot/pilot-env directory using your preferred version of Python:
cd '/install/path/of/gpt-pilot/`
rm -rf pilot-env venv
/path/to/correct/python3 -m venv venv
./venv/bin/python -m pip -r requirements.txt
(on Windows it's the same, except it's easier to remove the pilot-env folder in File Explorer, you need to use '' instead of '/' for paths, and you need to run 'pilot-env\Scripts\python.exe` instead of './venv/bin/python').
The Python virtual environment, which should have been automatically set up by VSCode installation, wasn't complete. You can do the same workaround as in the previous answer.
Yes! More info available in this article.
GPT Pilot currently gets slower the more files you have in your project. This is especially noticeable if you're using a framework that creates a bunch of files when initializing the project (eg. react, laravel, django,...).
GPT Pilot sends project files to the LLM. If you have large text files in your project, GPT Pilot might pick them up and trigger this error. TO avoid this, you can list those files (or folders) in fs.ignore_paths setting in your GPT Pilot config.json configuration file, like in this example:
{
...
"fs": {
...
"ignore_paths": [
... // default files that are already ignored by Pythagora
// add additional ignore rules here, for example:
"*.log",
"*.lock"
]
}
}
When you press the Send Database button in Pythagora, it doesn't send a database you've created with your projects. Instead, this button sends a separate log database containing your Pythagora LLM requests to our development team. This allows us to efficiently troubleshoot and resolve any issues you might encounter while using the platform.
Pythagora is pretty versatile and can handle building apps in any language. But for now, we're mostly doing our testing with Node.js, so we suggest sticking with using Pythagora for Node-based web apps. Our team’s still smoothing out the edges for web app development, but once we nail that, we’re planning to branch out to other platforms. Electron will likely be one of the first ones up.
With Pythagora Pro, you’re prepaying for usage, so it doesn’t expire. Your balance only goes down as you use it, but if you don’t, your money stays put in your account.
Thanks! The best way to help right now is to tell your friends about GPT Pilot, help spread the word about us, and be active in our Discord community!