It would be cool to provide instructions for running this in docker. I tried it yesterday and got it running although it gave an error when trying to use it. I did not look into what was causing it yet though.

Hi

this works great thank you!

WOW

im getting Invalid credentials in Authorization header

not really that familiar with running stuff locally

getting error as "Invalid credentials in Authorization header"

getting error as "Invalid credentials in Authorization header"

Are you sure you did those steps correctly?

1. Create a token with inference permissions: https://huggingface.co/settings/tokens/new?ownUserPermissions=repo.content.read&ownUserPermissions=repo.write&ownUserPermissions=inference.serverless.write&tokenType=fineGrained then copy it to your clipboard
2. Create a new file named .env in the Deepsite folder you cloned and paste your token in it so it should look like this:

HF_TOKEN=THE_TOKEN_YOU_JUST_CREATED

3. Launch the app again

verified steps, it launches but upon prompt I get the same response Invalid credentials in Authorization header

Hi guys, I gonna take a look at this, will keep you updated

@Vrajce @pedrod8 could you try again please ? Should be fixed now. (Please git pull first)

Using with locally running models would be cool too.

I thought it was free, but I get that message

When I click on the button, it tells me Invalid client_id

I've confirmed my token can login via huggingface cli and has inference access - I'm still getting the following error when i go to process any text

@pedrod8 Did you correctly git pull the changes ?

@diguishm could you git pull too and try again please ?

I did everything according to the steps above, it worked the first time. Thank you.

P.S..
Updated the node

it has started working for me now - many thanks!

Using with locally running models would be cool too.

I know right

I used the Dockerfile, set the HF_TOKEN and on first try get the error message: We have not been able to find inference provider information for model deepseek-ai/DeepSeek-V3-0324. Error happens in try catch when calling client.chatCompletionStream.

can i point this to my own deepseek API key and run offline using my own API key nothing to do with huggingface?

can i point this to my own deepseek API key and run offline using my own API key nothing to do with huggingface?

I did so after the previous error with the inference provider but always run into the max token output limit and receive a website that suddenly stops. Wondering how the inference provider approach works differently towards this.. i can not explain myself as deepseek is just limited to max 8k output.

can i point this to my own deepseek API key and run offline using my own API key nothing to do with huggingface?

I did so after the previous error with the inference provider but always run into the max token output limit and receive a website that suddenly stops. Wondering how the inference provider approach works differently towards this.. i can not explain myself as deepseek is just limited to max 8k output.

There is models now well over 1Mill so could easily swap. did you run the docker and set API and .env file?

Deepsite uses deepseek (here online) so this was the base of my test... Here online i receive a full website but locally with my direct deepseek api not. Yeah there are some models with much more output. also deepseek coder v2 with 128k .. but still wondering the differences between deepseek platform api and inference provider - makes no sense.

Can I run it locally without a PRO plan? If so, how do I set it up?

Yes you can use it without being PRO, but you're always concerned about limits (https://huggingface.co/settings/billing)

Thanks.

Can run locally - offline - with OLLAMA server

Hello, I subscribed to the pro option twice and they charged me $10 twice but I still haven't upgraded.

How to add google provider to this project? I want to use gemini 2.5 pro

Hello, after local deployment, here's the message I receive when I execute the request: "Failed to execute 'json' on 'Response': Unexpected end of JSON input."
How can I resolve this issue, please?

Hello, I subscribed to the pro option twice and they charged me $10 twice but I still haven't upgraded.

Very weird we are going to take a look at it (did you subscribe from hf.co/subscribe/pro?)

Hello, I subscribed to the pro option twice and they charged me $10 twice but I still haven't upgraded.

Very weird we are going to take a look at it (did you subscribe from hf.co/subscribe/pro?)

Yes of course via this link: https://huggingface.co/pricing, I was charged $20 for both tests

Can run locally - offline - with OLLAMA server

Which LLM model you use?

i use distilled DeepSeek, and Qwen 2.5 and Gemma 3

However I am sure to make this work i have to do something with code, but have no idea what.

Failed to fetch

i use distilled DeepSeek, and Qwen 2.5 and Gemma 3

However I am sure to make this work i have to do something with code, but have no idea what.

SO can we use it with this local setup or not?

Yeaa I was wondering the same thing, can we actually use a local LLM to run this or not!!!

hi i want to run it locally its my first time trying to run LLM locally can you provide me with step by step how to do it, what i need to install first what tools or software that i need thank you

please help to let me know, why I'm input the request, after AI running, the program always to display "极速赛车开奖结果历史记录”, also terminate the programming or let the display not correct, why happen this problem and how to input can let system not appear this problem.

Hell yeah thnx brother!

How do I share it as a website ? Like google slides

I created a custom version of DeepSite to run locally!

Now you can run the powerful DeepSite platform directly on your own machine — fully customizable, and with no need for external services. 🌟
Using Ollama, you can seamlessly integrate any AI model (Llama 2, Mistral, DeepSeek, etc.) into your setup, giving you full control over your environment and workflow.

Check out the project on GitHub: https://github.com/MartinsMessias/deepsite-locally

Does it only create front end content or will it actually function when you create something? I was trying to make a few thingd but I'm new to HF but I love making prompts with it. I just don't know how to turn it into something functional.

I got the same question as spiketop, can we make it functional?

If anyone knows lmk. DM or email me [email protected] because this is cool but I don't know how to make it work

It may not always add functionality beyond some hover and click effects, but I guess it depends on how you prompt it? I mean... It's a big model, you can go wild with your requests for it...

I also have the same question at Spiketop. Can I make website created functional? Anyone know how this can be done and willing to assist? Thanks!

Guys, one thing Im trying to do is to use the html code from the AI assistant and drop it on a VSC to make it usable. I was able to recreate the exact same page in there, and possibly making it functional.

Using with locally running models would be cool too.

that means: no money for they :/

I created a custom version of DeepSite to run locally!

Now you can run the powerful DeepSite platform directly on your own machine — fully customizable, and with no need for external services. 🌟
Using Ollama, you can seamlessly integrate any AI model (Llama 2, Mistral, DeepSeek, etc.) into your setup, giving you full control over your environment and workflow.

Check out the project on GitHub: https://github.com/MartinsMessias/deepsite-locally

I was thinking on the same thing.

i only can say, there is a huge diff between running locally and remote, obviously when running locally it depends of your possibilities, my posibilities are:

low GPU 4Gb
decent RAM 32 Gb
low CPU AMD Ryzen 7

the results are the next, when asking for a mobile UI for IA Chatting:
(stars are on my oppinion, the rate of results)

Gemma3 4B (LOCALLY) ⭐⬛⬛⬛⬛ 👇🏼

Gemma3 12B (LOCALLY) ⭐⬛⬛⬛⬛👇🏼

DeepSeek Coder V2 16B (LOCALLY) ⭐⭐⬛⬛⬛👇🏼

DeepSite (Default REMOTE model) ⭐⭐⭐⭐⬛👇🏼
(i know the interface is the one that runs models locally, but i recovered the output HTML from saved file, results of original deepsite repo)

@Usern123454321 Running models locally can get heavy fast. Using OpenRouter is usually way more practical. Models like Claude are affordable and perform really well. And DeepSeek V3 has been surprisingly good, especially for front-end tasks, easily one of the best in that area lately.

Good afternoon, I'm building a website for a client and in desktop and mobile formats it says "Made with Deepsite - Remix". I wanted to know how to remove it, because it doesn't look good. Can you tell me if there is a plan for what we can remove?

@antocarloss Its pretty straight forward to remove, unless if you vibe code it and dont know anything about programming. In that case, reach out to someone who can and pay them part of what you charge your client. It will literarily take less than 5mins to do so if you know what you are looking for.

How can l take the website from Deepsite hands

What is the minimum hardware requirements to run a V2 16B locally?

How do I run locality has been created

We have not been able to find inference provider information for model deepseek-ai/DeepSeek-V3-0324.

do this come with a model or i have to add my own model?

j'essaye de creer une application, comment faire pour passer le code de deepsite en appli????

I have followed the steps, and was able to install it as well. When I got o localhost:3000, it loads, but I don't see the Local Usage tag, plus I am getting invalid headers even when I have the .env file with the token. How do I resolve this? Thanks for the help!
I have confirmed that I have latest git.

Hello, after local deployment, here's the message I receive when I execute the request: "Failed to execute 'json' on 'Response': Unexpected end of JSON input."
How can I resolve this issue, please?

@web3gn did you manage to figure out this issue?

Is this all Available only on Linux?

The string did not match the expected pattern.

I Get this error

Criei um site como faço para copiar o link do site que criei

is it free if i run it locally? cause it's asking me for some pro subscription

How do we use the local ollama api or models for this project?

Hi there! 👋
I’ve made some changes and need to test them, but I’ve run into an issue — my free inference quota has been exhausted.
I'm currently exploring options to continue testing,
If anyone knows a quick workaround or a way to test without upgrading right away, I’d really appreciate your input!

what exactly are you testing?
send me the code, I'll try to run it on my paid version, I also made my custom development and connected an openrouter, everything works fine, but there is a problem that cannot be solved, this is the output of 1200-1700 lines of code, through any model, I tried many models, Claude 3.7 OpenAI 4.1 and many others, there is also a problem with the limitation in the number of tokens for 1 request (= 16 000), if you need to write large amounts of code and do many iterations.
Enzostvs is a very smart guy and he did a great job, there are clearly not enough video tutorials that will help with custom modifications of DeepSite

what exactly are you testing?
send me the code, I'll try to run it on my paid version, I also made my custom development and connected an openrouter, everything works fine, but there is a problem that cannot be solved, this is the output of 1200-1700 lines of code, through any model, I tried many models, Claude 3.7 OpenAI 4.1 and many others, there is also a problem with the limitation in the number of tokens for 1 request (= 16 000), if you need to write large amounts of code and do many iterations.
Enzostvs is a very smart guy and he did a great job, there are clearly not enough video tutorials that will help with custom modifications of DeepSite

I was testing some custom changes I made to DeepSite’s interface — mainly added features like file upload support, conversation history tracking, and a microphone input option. Everything works smoothly on the frontend, but now I need to validate backend inference after these changes.

I’ve hit a limit on my free inference quota, so I can't test full cycles right now. Since you're running a paid setup and even connected OpenRouter, that's awesome! If you're open to testing, I’d be happy to share the code (just let me know how you'd prefer I send it).

Also agree — not enough clear tutorials out there for custom DeepSite setups. Maybe we could even collaborate on a guide

.

what exactly are you testing?
send me the code, I'll try to run it on my paid version, I also made my custom development and connected an openrouter, everything works fine, but there is a problem that cannot be solved, this is the output of 1200-1700 lines of code, through any model, I tried many models, Claude 3.7 OpenAI 4.1 and many others, there is also a problem with the limitation in the number of tokens for 1 request (= 16 000), if you need to write large amounts of code and do many iterations.
Enzostvs is a very smart guy and he did a great job, there are clearly not enough video tutorials that will help with custom modifications of DeepSite

I was testing some custom changes I made to DeepSite’s interface — mainly added features like file upload support, conversation history tracking, and a microphone input option. Everything works smoothly on the frontend, but now I need to validate backend inference after these changes.

I’ve hit a limit on my free inference quota, so I can't test full cycles right now. Since you're running a paid setup and even connected OpenRouter, that's awesome! If you're open to testing, I’d be happy to share the code (just let me know how you'd prefer I send it).

Also agree — not enough clear tutorials out there for custom DeepSite setups. Maybe we could even collaborate on a guide

@sayasurya05
Can you explain why your code that you sent me in telegram asks for account data, login and password when trying to send the code to https://huggingface.co/
maybe it's worth making the code public, as it does for example https://github.com/MartinsMessias/deepsite-locally ???

FEATURES.md

Enhanced Features for DeepSite

This document outlines the enhancements made to the DeepSite application, focusing on the chat interface, search bar, and file upload functionality.

Chat Interface Enhancements

The chat interface has been completely redesigned to provide a better user experience:

• Chat History: Added a collapsible chat history panel that shows all previous interactions between the user and AI
• Message Timestamps: Each message now displays the time it was sent
• Message Status: Messages show their current status (sending, sent, error)
• Visual Distinction: Clear visual separation between user messages and AI responses
• Scrollable History: Chat history is scrollable for easy navigation through past conversations
• Clear History: Option to clear chat history when needed

Search Bar Improvements

The search bar has been enhanced to support longer text inputs and provide a better user experience:

• Auto-resizing Textarea: The input field now automatically resizes based on content
• Scrollbar Support: When text exceeds the maximum height, a scrollbar appears
• Clear Button: Added a button to quickly clear the input field
• Keyboard Shortcuts: Press Enter to send, Shift+Enter for new line
• Improved Placeholder: Dynamic placeholder text based on conversation state
• Visual Feedback: Better visual feedback during input and when AI is processing

File Upload Feature

A new file upload feature has been added to allow users to share files with the AI:

• Multiple File Support: Upload multiple files at once
• File Type Filtering: Support for HTML, CSS, JavaScript, and image files
• File Preview: Visual preview of uploaded files with appropriate icons
• File Management: Options to remove individual files or clear all files
• Size Limitations: 5MB maximum file size with appropriate error messages
• Integration with AI: Uploaded files are sent to the AI for processing
• Visual Indicators: Badge showing the number of uploaded files

Server-Side Enhancements

The server has been updated to support these new features:

• File Processing: Server now processes uploaded files and includes them in AI prompts
• Content Extraction: Extracts and formats file content for the AI
• Image Handling: Special handling for image files
• Code File Support: Special formatting for HTML, CSS, and JavaScript files

How to Use

Chat History

1. Click the "Show Chat" button at the top of the chat interface to view chat history
2. Scroll through past messages
3. Click "Clear Chat History" to remove all messages

Enhanced Search Bar

1. Type your message in the input field
2. The field will automatically expand as you type
3. Press Enter to send or Shift+Enter for a new line
4. Click the clear button (X) to quickly clear the input

File Upload

1. Click the attachment button (paperclip icon)
2. Select files from your device (HTML, CSS, JS, or images)
3. View uploaded files in the file panel
4. Remove individual files or clear all files as needed
5. Send your message with the attached files

Technical Implementation

The enhancements were implemented using:

• React state management for chat history and file uploads
• LocalStorage for persistent chat history
• React-textarea-autosize for the expanding input field
• WebSocket integration for real-time updates
• Server-side file processing with Base64 encoding/decoding
• Tailwind CSS for responsive design

IT WOULD BE NICE IF YOU MADE IT CONVERSATONAL, AND NOT JUST CODE BECAUSE IT IS NOT FOLLOWING INSTRUCTIONS, I BUILD A WEBSITE AND ASKED FOR A CHANGE IT CHANGED THE ENTIRE SITE AND I CAN'T GET BACK TO THE ORIGINAL SITE, I AM NOT A DEVELOPER, I LOVE THIS BUT IF YOU COULD MAKE IT CONVERSATIONAL SO I CAN TALK TO IT WITH LOGIC THAT WOULD BE GREAT! (IF THAT MAKES SENSE, LOL AGAIN I AM NOT A DEVELOPER, I'M STUDYING BUT I DONT UNDERSTAND CODING YET)

ALSO MAYBE IF WE COULD HAVE A TOGGLE TO GO TO A PREVIOUS VERSION? LIKE A HISTORY TOGGLE THAT WAY IF IT CHANGES SOMETHING WE DIDN'T WANT CHANGED WE CAN GO BACK TO THE PREVIOUS VERSION AND START FROM THERE

Hi @batesbranding good idea
Just implemented it, it's a state only history. Once you refresh the page, the history is clear.