can we add a http proxy?

This should be fixed now - code is always aligned to the left, and the column is limited so it won't grow too large.
Originally posted by @Niek in #13 (comment)

Reopening #13

I think there are still problems with this issue. The user text in general is right-aligned, which make it really hard to read the code you have written in a previous message. I know the "right" solution is for a user to use a code block, but I don't think most people do that when copy-pasting code into ChatGPT

I got this when I run it :
VITE v4.1.4 ready in 557 ms

➜ Local: http://localhost:5173/
➜ Network: use --host to expose
➜ press h to show help

How to expose to other computer on the network?

I tried npm run dev --host , but it's not work

May I ask: is it with model GPT-3.5-turbo?

Possible to add Copy button on code block?
Thanks!

Hi! Great project, currently looking into it. I have tow questions though. Is there a setting to toogle voice reply? For example I want to send a message using my voice, but I don't want a voice reply. Also, how to stop a voice reply once started or change it's speed?

How can I choose to interact with a specific fine tuned model that I trained?

The chat content contains code, but it appears blank on the screen. Is it a filtering issue?

I'm self-hosting and would rather the API Key wasn't visible on-screen every time I launch; could it be collapsed by default?

I think it would be great if I could set my API key as an environment variable in .env so that I can save the effort of entering it at startup.

Hi,

I think the possibility to edit a message like you can do on ChatGPT would be an amazing addition to the project.
To do this it looks like you might need to change the storage of chat from an array to a tree-like structure. And then of course change the API call and chat display parts as well.

I might be able to look into it in a couple of weeks if no one else gets around to doing it first.

Best

It'd be helpful if we could rename chats.

It'd be amazing if chatgpt-web could name the chats automatically like ChatGPT does (I assume it's making a separate call to the model asking for a summary of the topic being discussed in the user's opening message?)

Hi there,

I have a concern that how to protect this service from unauthorized access when I expose it on the Internet.
Could you give me some advices?

Thanks
Jarvis

Add the possibility to delete a chat

ChatGPT-Web: Customizable: You can customize the prompt, the temperature, and other model settings. Multiple models (including GTP-4) are supported.

API:
I just downloaded the updated Chatgpt-web and have run the localhost.

On the demo site, in settings there are multiple GPT APIs to select from. (GPT-4, GPT-4 32k, GPT-3, etc.)

But when I run locally there are only two GPT APIs to select from only GPT-3 is available in settings.

The instructions say that it is customizable, but it is unclear how to add the other APIs so that they can be selected from.

Settings
In Chat settings, there are a list of settings, but it is unclear what the settings do or how to set them.

Prompt
How do I add and remove my own reusable customizable prompts for easy access?
How do I remove prompts from default awesome-chatgpt-prompts list I don't want but keep what I do want?

src/taurus
It is unclear how to launch the desktop version and what that launch would look like.

Unknown Features/Feature Suggestions
Below are unknown features - not sure if the webui has these features. If they do not have these features I would like to suggest them - if no plans for them, I would like to find ways to implement - will you point me to the files to add new features if not available.

Theme:
The Github page shows a black theme, but it is white on local. How can I change to preferred theme?

Chats Exports
Is there a way to export all chats at once (instead of one at a time)?

Storage
Right now storage is set up in a temp file of browser setting. Is there a way to change this so each chat is saved locally without need to export to move between folders.

Chats
When too many chats is there a way to have the list scroll instead of making the page grow bigger?

Versioning
Does your ChatGPT-web utilize versioning - I have an older version of Chatgpt-web with less features than this one, but it is unclear which version it is in comparison to this one.

Unsure how to know when there is a new version update without thinking to check the page, will check frequently. Is there an easy command to check for and update the webui via cmd? Or is it better to manually check and download new zip in new folder and launch from there with each update?

The app has improved so much since the first iteration! Thank you friend!

Just like OpenAI's playground
https://platform.openai.com/playground?mode=chat

Downloaded the local version and did the docker-compose up -d, not sure how to follow up to enable GPT4

When code is generated from ChatGPT, it is clearly rendered as a code block. But syntax highlighting doesn't seem to be working in that code block. I do see relevant code in the project for syntax highlighting but perhaps the text color got overwritten with some CSS styles?

I was surprised there were none. Those are by far the most useful premade prompts!

This app is great. Thank you for your wonderful work!

When I’m typing in ios, the window keeps trying to slide up, which makes the app painful to use on my phone. I’m guessing it’s related to the focus of the window, since there’s a selected item up in the chats tabbar.

It seems that Line 420 in Chat.svelte

<article class="message is-success assistant-message">

in order for the following specification in app.scss to be effective:

.assistant-message {
  max-width: 90%;
  width: fit-content;
}

It would be nice to sync with this repo https://github.com/f/awesome-chatgpt-prompts and get access to a list of pre-written prompts from a list

I'm super impressed with https://mlc.ai/web-llm/ - would be interesting to see if and how we can integrate this in ChatGPT-web.

Repo: https://github.com/mlc-ai/web-llm
Demo: https://mlc.ai/web-llm/#chat-demo

CHATGPT is spelled CHATGTP in index.html

I'm encountering an issue with formatting on multiline messages from that contain wide code blocks. As a user message, the text gets right-aligned off the screen (also, I can't scroll left or right, so I have to use the tab char to highlight the 'Clear Messages' button in order to get the viewport over to the right)

I have included example input below (I'm passing the markdown source as my chatgpt input).

Having the code blocks have a max size & scroll any overflow might be a good solution here, although another more general problem this brings up is multiline user input with code blocks isn't a great match for the left/right aligned styling of messages

Example Input

Please generate SQL for my queries. Examples:

Example 1: Select all ABC with ID 1,2,3:

select abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx, abc.xxxxx
from schema.some_table abc
	     left outer join ...
where abc.id in (1,2,3);

Some text after the snippet

First of all, thanks for your work on the project!

I believe setting an "init" prompt would be interesting so that the chat is already set up for a specific task.

Hi, just came across this when looking for a ChatGPT API UI, and this looks like a great implementation.
Currently, while the source is exposed/available in this repo, I could not see any kind of license or grant of rights so no actual grant of usage/modification/distribution of the code is currently granted (and hence is not currently typically open source, unless I've missed the license).

Do you intend for this to be Open Source (open grant of usage, modification and distribution)?

This message was generated on gpt-3.5-turbo-0301 using 179 tokens

Both the default and the set value are gpt-3.5-turbo, yet the prompt still shows gpt-3.5-turbo-0301

now we meet 4096 token limit, maybe we can drop the oldest message when its get the limit?

It is light mode set as default theme, I would like to switch to dark mode the same as it was introduced in the home page.

Would you share how to make it ? Thank you.

Display the total amount of money consumed by this API next to the input field of API

When I open the tool, I always land on the "enter an api key" page - we could start typing with one click less, if the tool would check if there already is a token and then directly show the "new chat" view

Should we use Prettier to standardize the formatting of the code?

I tried to set the default model for new chats in file Chat.svelte, but new chats always default to "gpt-3.5-turbo" despite me changing it to 'gpt-4' in file.

Chat.svelte snippet:
const modelSetting: Settings & SettingsSelect = {
key: 'model',
name: 'Model',
default: 'gpt-4',
title: 'The model to use - GPT-3.5 is cheaper, but GPT-4 is more powerful.',
options: supportedModels,
type: 'select'
}

Hiya,

I downloaded the zipped file of the code, used sftp to upload on my server but when I load the folder name, I just get a blank page. I know I am definitely missing some steps or doing something wrong but can't seem to figure out what I need to do in addition

Extra info - I use cloudways for hosting my web server

why did not set the response word by word

As far as I know, ChatGPT prepends a "system" message to the conversation as a guideline for the assistant.

In their documentation they mention this example:

import openai

openai.ChatCompletion.create(
  model="gpt-3.5-turbo",
  messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Who won the world series in 2020?"},
        {"role": "assistant", "content": "The Los Angeles Dodgers won the World Series in 2020."},
        {"role": "user", "content": "Where was it played?"}
    ]
)

We could perhaps allow users to specify a system message or to provide a list of prompts to choose from depending on the task (e.g. assistant vs translator vs help with summarisation)

Hi, I really like this project and I have several suggestions:

1. Is it possible to support latex rendering in the browser? For example, equations like $H^2 = \frac{8\pi G}{3}\rho - \frac{k}{a^2} + \frac{\Lambda}{3}$ could be displayed in a much more elegant way. I noticed this equation can be rendered on GitHub due to "Styling with Markdown is supported"; maybe that's the easiest option for implementation?

2. Is it possible to add a button to change the default background color between white and black?

Thanks!

It would be awesome to be able to switch between different models. I usually go with gpt-3.5-turbo and text-davinci-003 on my python gui. Depending on what I look for, I sometimes switch back and forth between them in one conversation.

Implementing the option for gpt-4 would be great for those with early access. I've also come across the codex-davinci model which looks promising, although I did not have great success with it.

Using horizontal sliders instead of the small increment up/down arrow buttons in the settings section would make it a lot easier to use. Additionally, providing min/max value for the different settings would make navigation through that section more user friendly.

Thanks, this project is amazing, keep up the good works!

Hey 👋 !

Awesome project!

I'm trying to run chatgpt-web with llama.cpp, I've created a project using golang lama.cpp bindings https://github.com/go-skynet/llama-cli which mimics the OpenAI API to be 1:1 compatible but having multi-model that can run locally instead.

It seems all to work so far, and I'd like to document the usage to have them both working together, so to use it with local models. However, I'm struggling as chatgpt-web seems to filter the model from the API with openAI available models - llama-cli returns a list of models but the filtering chatgpt-web is doing prevents to select models from the list. (e.g. alpaca can't be run unless I do some hardwiring on the API).

If you want to test it, you need to run llama-cli from the latest image built from master, like so:

./llama-cli api --address 0.0.0.0:8080 --models-path models-path-here --threads 14

And set the VITE_API_BASE accordingly in the .env file.

It would be super-cool if could work together to have the capability to load local models, maybe directly adding options to run it aside with docker-compose (that's what I'm currently doing!) WDYT?

Err msg:
SyntaxError: Invalid regular expression: invalid group specifier name

I found that it may relate to this issue in library flourite
teknologi-umum/flourite#72

Thanks in advance for fixing it...

I used it with Lencx desktop app as a dashboard it works pretty well
I know it's very basic, but can we have Dark Mode support I would really appreciate it

Many people in my country can't use ChatGPT, and to help them use it, I'd like to make some changes based on your code.I hope that instead of entering a KEY, the user requests my built-in KEY。

Can you give some suggestions, or can you come up with a copy of this version of the code?

Thanks!

When using ChatGPT, I often start with a query to contextualize the question, then proceed to ask several things pertinent to the original topic. As the conversation grows, there two main issues:

• Each new query becomes more and more expensive
• The original context is getting "poisoned" with the model replies and it's also getting further back in the conversation, likely weighting less in the next predictions.

The proposed "advanced" feature proposed is the ability to manually select what messages are used in the next query.

Very simple example:

1. User: Conversation about president Barack Obama. When was he born?
2. AI: He was born in XXXX
3. User: What did he graduate in?
4. AI: He studied XXXX
5. User: When did he become president?
6. AI: In XXXX

Here 3 and 4 are not needed to answer 5 and could be manually switched off with e.g., a checkbox that is only showed if an "advanced features" setting is turned on.

Now that we have the option to edit messages, we can regenerate responses by just using editMessage on the last message while keeping the same content.

I noticed that when I set the model back to gpt-3.5-turbo, the info text still thought I was using gpt-3.5-turbo-0301:

This message was generated on gpt-3.5-turbo-0301 using 1727 tokens ~= $0.003454

When I checked the console, the request body still had the "model" field as gpt-3.5-turbo-0301.

Would love to see this support Azure OpenAI API!

Currently there's no way to send messages that contain newline chars. It would be useful to be able to paste in pre-formatted messages, and to use shift+enter to add a newline within the current chat message