Feature: Advanced OpenAI configuration

[sebiweise]

In this PR I´ve added some OpenAI configuration parameters, updated the .env.example file and did some package updates for NextJs, OpenAI and Prisma.

OpenAI can no be configured via ENV vars. The following variables were added:

VAR name	Default	Description
OPENAI_API_URL	https://api.openai.com/v1	This parameter can be used to define a custom OpenAI endpoint. I´ve tested this with the official OpenAI endpoint, Ollama and also with a Cloudflare AI Gateway
OPENAI_CATEGORY_EXPENSE_MODEL	gpt-3.5-turbo	This parameter can be used to define a different OpenAI model that will be used for the expense category determination
OPENAI_CATEGORY_RECEIPT_EXTRACT	gpt-4-turbo	This parameter can be used to define a different OpenAI model that will be used for the receipt extract feature

Using the Cloudflare AI Gateway will help decrease the amount of request that will be send to OpenAI and therefor decrease the cost that will be generated. Same expense titles for exampel will be cached by Cloudflare and the can be reused for the next requests.

It is also possible to define Ollama as your AI endpoint. Ollama is compatible with OpenAI, therefor the OpenAI SDK´s can be used with Ollama (Docs).
Just set your Ollama server url as your OPENAI_API_URL="http://localhost:11434/v1" and define your preferred local model in OPENAI_CATEGORY_EXPENSE_MODEL and OPENAI_CATEGORY_RECEIPT_EXTRACT.
OPENAI_API_KEY cannot be null and must be set to some random stuff, for Ollama it won´t be used.

[ChristopherJohnston]

See #166

[sebiweise]

See #166

Yeah as I see so far, your PR and your changes aim to add config options for Ollama and not to change anything OpenAI related.

[ChristopherJohnston]

No worries, I was just linking them as they are related and there may be some merge conflicts for the other if/when one is merged.

[sebiweise]

No worries, I was just linking them as they are related and there may be some merge conflicts for the other if/when one is merged.

Alright, but I think we can skip the use of fetch and just use my change to also support Ollama. As far as I can understand just changing the url and the model will do the trick for Ollama.
Have a look at this documentation of Ollama: https://ollama.com/blog/openai-compatibility
Ollama should be OpenAI SDK compatible🤔

[sebiweise]

Yes I can confirm, that you can use my changes to use Ollama local models like this example:

OPENAI_API_URL="http://127.0.0.1:11434/v1"
OPENAI_CATEGORY_EXPENSE_MODEL="llama3"
OPENAI_CATEGORY_RECEIPT_EXTRACT_MODEL="llama3"

[sebiweise]

Updated the PR description to include the Ollama instructions.