+1 on Add RAG (and web search)

+1 on Add RAG (and web search)

+2 🔥

Add function calling capability

Two features that would make assistants infinitely more useful:

1. Function Calling - Capability w/User Defined functions endpoints. (ie. user could provide function calling schema + REST endpoints for function calling)
2. RAG - Ability to upload documents (PDF, txt, URLs) that are chunked and stored into a VectorDB for RAG with the assistant.

How do Assistants work under the hood? Do they use OpenGPTs from Langchain?
Happy about a link to the docs or something

How do Assistants work under the hood? Do they use OpenGPTs from Langchain?
Happy about a link to the docs or something

Ok just checked out the source code, doesn't look like OpenGPTs but their own thing

Amazing!

Amazing! I created an Assistant but it isn't showing up on the list, does it only show popular ones?

It's great, I love this community. I created my GPT and it is working great.

I hope that in the nearer future it can be added to inclusion topics for those people who have some physical difficulty such as: vision, listening and speaking, they can use these technologies.

It’s amazing, would be great to be able to upload long documents similar to OAI GPTs, maybe using something like MemGPT?

What are the licenses on the actual Assistants created in
https://huggingface.co/chat/

Can I copy some of the assistants into my own hosted version of chat-ui?

If there are unknowns, what about at least the assistants created by the huggingface team? Can I use these? Since most of these starter assistants are created by members of huggingface.

It would just really help if I can at least get a couple of these starter ones.

A stretch goal, or further extension: Can there be a tool to let others import the assistants created under huggingchat, into our own chat-ui implementation? That would just be awesome!
Without it, it is like we are now all missing this amazing feature, and possible have a fragmented community if we ever have our own chat-ui.

thanks!

Add RAG to Assistant is surely a big deal for me.

On my own hosted version of chat-ui, a main added feature is my own RAG which I call "knowledge base" (it works the same as web search, just with RAG). Now if I can make use of this assistants feature to add my own RAG, that would be amazing!!! In fact, that is the most important next feature for my product.

thanks!

How do Assistants work under the hood? Do they use OpenGPTs from Langchain?
Happy about a link to the docs or something

Ok just checked out the source code, doesn't look like OpenGPTs but their own thing

How do Assistants work under the hood? Do they use OpenGPTs from Langchain?
Happy about a link to the docs or something

For now, it looks like its just a custom prompt. I hope RAG and Function Calling will come next.

This is fantastic! Definitely would like to be able to adjust the temperature. Asking the assistant "how are you?" gives 3 paragraphs of spiel.

Thought I lost my assistant.

Here it is. It is called "Coder". In case I lose it or you want to try it. I hope other functions are also added that are mentioned by other users.
Another Assistant I created is "Learn English With AI" here.

100% add RAG. Also, there should be an option to hide our custom instruction in the setting.

When can we have API for assistants?
When are you planning to expose on Vercel AI SDK ?

is Assistants on HuggingChat opensource?

is Assistants on HuggingChat opensource?

Yes, https://github.com/huggingface/chat-ui

Will be so much bots name on future, you can add search name bots.

Consider adding category tags for easier searching

Hello, is it possible to add a search engine? to filter by the names of the created attendees.

I feel bit nude to share my own Bot System Prompt. I think it's good when Creator bot can choose to hidden their own System Instructions.

is it possible to save these customized models to the model hub?

Image generation would be a nice touch.

can the created assistant be available as REST Endpoint to access outside of hugging face ?

I feel bit nude to share my own Bot System Prompt. I think it's good when Creator bot can choose to hidden their own System Instructions.

isn't that against the purpose of open-source? I mean, for now the used techniques in the system-prompt are the only interesting thing about current assistants. to me, usage of assistants with their current abilities will be not of any direct value - the only interesting thing is the way people formulate their instructions. IMHO hiding these instructions, reduces the value of the assistants to zero.

I feel bit nude to share my own Bot System Prompt. I think it's good when Creator bot can choose to hidden their own System Instructions.

We are thinking about whether or not to add an option to "hide" the prompt.

one of the reasons we are unsure for now (in addition to the spirit of open source and knowledge sharing as @LHind-RobG says) is that no LLM prompt is ever really "hidden" ie. any attempt at preventing retrieval is a bit hackish/non robust.

I created 3 assistants but only see the one I asked the last question of, how can I see all of my assistants that I created

I created 3 assistants but only see the one I asked the last question of, how can I see all of my assistants that I created

@Paul-Arista here: https://huggingface.co/chat/settings in the list on the left.

Given that we know that people remake diseased partners or loved ones in ai to talk to them again or throw confidential company structures/documentation into system prompts etc, is it really a good idea to enforce that every assistant is to be open and accessible to everyone?
Its obviously not great that company (or realistically trade-) secrets and very private personal information get thrown onto huggingchat servers but its a lot better than having that available to everyone who cares to look.
Since everything is open source I suppose you could just run everything local, but (as far as I can tell) I dont have the hardware and barely enough tech knowledge to set that up and I doubt most people even have that much.
But Im also not sure how you'd then prevent selfish use of everything here with the ability to make assistants private to the user who created them.
This is pretty tough and Im not sure there even is a good solution here, to be honest.

如何使用api 調用個人助理？

what he said was 'How to use the API to call a personal assistant'

what he said was 'How to use the API to call a personal assistant' 抱歉，我以為這裡有自動翻譯。
感謝您幫忙。

… is that no LLM prompt is ever really "hidden" ie. any attempt at preventing retrieval is a bit hackish/non robust.

All of the ChatGPT system prompts, the "very secret" Bing AI prompt, etc. have been leaked.

@guary I mean just opt-out from sharing your conversations and don't use assistants feature? You can use system prompt without making it public assistant.
Also you could check huggingchat privacy policy if you are really concerned about privacy:

"By default, your conversations may be shared with the respective models' authors to improve their training data and model over time. Model authors are the custodians of the data collected by their model, even if it's hosted on our platform.

If you disable data sharing in your settings, your conversations will not be used for any downstream usage (including for research or model training purposes), and they will only be stored to let you access past conversations. You can click on the Delete icon to delete any past conversation at any moment."

1. What will happen to the Assistant if a model is taken down? Migrate to new llm with context token +prompt as we/bot authors can change sys prompt of the assistants anytime? unlikely ig. or we could have a migration system for our old chats.
2. shall there be a "View Sys Prompt" just like in regular chats beside/below the bot button? As the assistant button at the top shows the latest prompt only while the chat might have started with another prompt. (doesn't change the already active chat really)(once it recognized the changed sys prompt upon me mentioning only a part of it)

I'm getting "must be logged in", though I am logged into Hugging Face, when trying to create a new assistant. Is there some additional way I need to log in?

I'm getting "must be logged in", though I am logged into Hugging Face, when trying to create a new assistant. Is there some additional way I need to log in?

No that might be a bug but I cannot reproduce, do you see your username in the bottom navigation?

Je sais que la vie n'est pas si évidente que ça pour tout le monde mais étant dans le domaine de l'informatique si je peux vous être utile voir même participer financièrement pour que ces fonctionnalités restent gratuites et à porter de tout le monde : vous pouvez compter sur moi ! Et vivre le savoir et la technologie libre et pour tous !
Merci la communauté ❤️

More features that would make assistants infinitely more useful for ( non coders )

Auto Function Creator: A simple prompt could lead to the creation of custom functions.
It's about making programming accessible to all, empowering creativity without deep technical expertise

@victor It was properly bc I started in guest mode, but then couldn't find a way to leave that mode. After trying things out for some time a pop-up to log in appeared, and then I could solve my problem.

Je sais que la vie n'est pas si évidente que ça pour tout le monde mais étant dans le domaine de l'informatique si je peux vous être utile voir même participer financièrement pour que ces fonctionnalités restent gratuites et à porter de tout le monde : vous pouvez compter sur moi ! Et vivre le savoir et la technologie libre et pour tous !
Merci la communauté ❤️

@Batwilf HuggingFace est une boite avec un fond de +4 Milliard dollars 🤯 Mais il y a tjrs des communautés et des groupes qui travaillent dur et finallement publier leur traveaux open source et je pense que la majorite (meme tout le monde) ont tjrs besoin de ce type de support 💪🏻🔥 Merci pour ton engagement envers Open Science 🤗

How much text can you put into the custom instructions?

i have one request that might be a bit of a reach but would be an awesome feature. if there could be a way that we can make an api call or some other call out to a custom model, we could use the model for other applications.

I think it would be useful to be able to attach content to ground the assistant like PDF, word docs, spreadsheets etc.

Pls add (1) memory and (2) another layer with another model (or the same), to moderate, or to direct the first one (who can actually talk to users)

Can we change the preview of an assistant for social media posts? It looks too much like another website's. Just a background color change would do the trick.

Do these models have guardrails built in already? Or do we need to include them in the prompt?

Do these models have guardrails built in already? Or do we need to include them in the prompt?

Depends on the model. For example, CodeLlama will have many guardrails, Nous Mixtral will not.

Do these models have guardrails built in already? Or do we need to include them in the prompt?

Depends on the model. For example, CodeLlama will have many guardrails, Nous Mixtral will not.

Mistral 7b? I’m only asking bc I haven’t been able to determine whether the models for the Assistants feature specifically include guardrails or not. Apologies, I wasn’t clear before.

Will there be a search feature for assistants?

Allow to create private assistants.

is there any API? What features do you need so that it can run on your PC?

is there any API? What features do you need so that it can run on your PC?

Yes, check this: https://huggingface.co/chat/settings

Is there any api we can use to call from like a python script though?

People, today, February 22, I have been asked several times, the GPTs are still processing the response and the model I am using is mistralai/Mixtral-8x7B-Instruct-v0.1, could it be that many requests are being made to the server and not Is it enough? help !

It would be nice to be able to make the assistants private

Is there a feature that allows the assistants to be categorized? for example, I'm interested in health topics, is there a way to classify them or search the assistants?

+1 Being able to change model params for assistants (temperature, repetition penalty, ...)

Does the assistant read&train on links in the instructions?

I'm getting this error now with most of the assistants:

Jsut getting started with this - ported my Custom GPT from OAI to here as this is more public / open ;)

This Assistant is a group of 4 stellar startup founders, go ahead and talk to all 4 of them ;)

https://hf.co/chat/assistant/65f46eb3896fca19f055ba0c

Can we have private Assistants so that we can experiment and test them before making it public?

Can we have private Assistants so that we can experiment and test them before making it public?

If you don't share them them they should not be visible anywhere to others.

Can we have private Assistants so that we can experiment and test them before making it public?

If you don't share them them they should not be visible anywhere to others.

When opening the window to create an assistant, a note at the top states that 'All assistants are public'. Not completely sure how it can be made private like that.

è possibile integrare un assistente chatbot personalizzato all'interno di una piattaforma cloud basata sul modello di macchina a stati che consente già di creare processi digitali personalizzati correlando singoli servizi tra loro?

Hi Victor. I made a chatbot about dictionaries, this is the link https://hf.co/chat/assistant/65c570eedf1b8733c8fc794b. It´s public but I do not see it when I browse the Community Chatbots. What could have happened ?

Hi Victor. I made a chatbot about dictionaries, this is the link https://hf.co/chat/assistant/65c570eedf1b8733c8fc794b. It´s public but I do not see it when I browse the Community Chatbots. What could have happened ?

@victor @nsarrazin Same issue for my assistant , https://hf.co/chat/assistant/6612cb237c1e770b75c5ebad

@KingNish @Flos0005 @flexagontnt it's like unlisted. unless you share the link and few people use it, it starts to appear in the assistants page ig
@victor although i do have a test assistant here assistants?user=svcy but it doesn't show from incognito/where i'm not logged in. I assume it's only a filter (by username) for assistants page+own bots? Thus, others' bots aren't visible here. but shouldn't they be available as it is said that all assistants are public by default?
it'd be best if we could have some sort of toggle option maybe..
and thanks anyway 😊

How can I make my assistant(https://hf.co/chat/assistant/65fc7e89c0f4abe9abb9d304) come to the Community list? I have reached 10+ users, but it's not showing up in it.

Maybe they only add featured bots?

@KingNish @Flos0005 @flexagontnt it's like unlisted. unless you share the link and few people use it, it starts to appear in the assistants page ig

@SvCy Can you do me favor, By checking out my assistant.
Assistant link - https://hf.co/chat/assistant/6612cb237c1e770b75c5ebad

@KingNish i already did. It's the fairly popular pollinations.ai. for now I can recommend that the numbers of generations should be lowered a bit or not to auto gen it's suggested ideas unless users confirm.. but it's your bot ofc.. so, whatever you think is good for it 🙂

@raduwolf Your amazon assistant is good, But the link it generates never works.

@Flos0005 Your dictionary is very good, it translates any word from any language. Super.

@KingNish i already did. It's the fairly popular pollinations.ai. for now I can recommend that the numbers of generations should be lowered a bit or not to auto gen it's suggested ideas unless users confirm.. but it's your bot ofc.. so, whatever you think is good for it 🙂

@SvCy Thank you for your Review. As you know Image generated by ai is not Perfect every time. So, I come up with idea to generate many images at once, so user do not waste their time and get best result.
See this example- https://hf.co/chat/r/ecanPCM
in this you can see some images have error. so, when you generate more images, you get images with more accuracy and also give options to choose. + Main motive of this ai is bulk generation.
But if you want to gen 1 or 3 images you can include it in prompt.

As you say to stop autogen suggesting idea. So , I will stop generating suggest ideas images, but suggesting ideas will still generate without image.

Saurav Thank you Once again for your detailed review.

Edit - Okk I updated AI if you give detailed prompt than AI will generate 1 image but if you give low quality prompt then ai refine its prompt with promptify and break low quality prompts into 8 good quality prompts and generate 8 images.
Also, from now more ideas does not generate any mage until you say it to generate.

@KingNish You're welcome!😊 if possible I'll engage more in your thread at #399. cya then!

• Improve assistant creation for beginners (maybe with a step-by-step wizard or an assistant that creates other assistants)
• Generate your Thumbnail Assistant via AI

@victor I am working on the assistant which help to create assistant step by step. Can you please check it or help in modifying it.
And is there any documentation or something which help it in improving.

https://hf.co/chat/assistant/66142eaa308c1b934f922a31 (Currently in alpha stage so, lots of error currently working to solve)

Feature That was added to it

1. Help in making assistant step by step
2. Help in Choosing write model and setting.
3. Help in generating System Instruction.
4. Help in Generating Thumbnails for assistant.
5. Help in solving errors while making assistant.
6. Help in Suggesting feature for assistant.

and here is example for making an amazon helper assistant https://hf.co/chat/r/rGHqcL3

Click Me for Example

And please suggest feature and bugs.

Interesting @KingNish - This makes me think we should add a transportable JSON schema for assistants cc @nsarrazin

• Generate your Thumbnail Assistant via AI

Feature added with promptifier 0.5 but result are okk okk 2/10

Edit: More improved version of both logo and instruction generator

I created logo of some Assistant

Results are Good from their current. 9/10

Some more logo

@julien-c Clone of Hugging Face CTO -

@DavidMcKay Professor GPT -

@nirajandhakal Coder assistant -

@KingNish Hugging Tutor

Interesting @KingNish - This makes me think we should add a transportable JSON schema for assistants cc @nsarrazin

Finally Done

@victor Check it - https://hf.co/chat/assistant/66142eaa308c1b934f922a31

Feature That was added to it

1. Help in making assistant step by step
2. Help in Choosing write model and setting.
3. Help in generating System Instruction.
4. Help in Generating Thumbnails for assistant.
5. Help in solving errors while making assistant.
6. Help in Suggesting feature for assistant.

# Request of Feature

Assistant of the week
It helps new assistant to reach people.
Just add it at Top (As like spaces.)

@victor Can you add article here about Hugging Chat https://huggingface.co/blog
Because there is no information about Hugging Chat.

Because I read many articles about Hugging Chat which are half true and not showing his full capabilities.

https://www.jagranjosh.com/general-knowledge/amp/what-is-huggingchat-is-it-better-than-chatgpt-1684146939-1

https://medium.com/@sudarshan-koirala/introducing-hugging-chat-assistant-build-your-own-personal-assistant-similar-to-openais-gpts-915d672285f8

https://www.zdnet.com/article/what-is-huggingchat-everything-about-the-new-open-source-ai-chatbot/

How do I adjust the repetition penalty? Every time I adjust it even a little it says “enter a valid value”

How do I adjust the repetition penalty? Every time I adjust it even a little it says “enter a valid value”

Place value between 0.8 to 1.2 till 1 decimal place.

Place value between 0.8 to 1.2 till 1 decimal place.

@KingNish where are we talking about placing them? running locally? or specifc spaces?

How do I adjust the repetition penalty? Every time I adjust it even a little it says “enter a valid value”

Place value between 0.8 to 1.2 till 1 decimal place.

It still doesn’t work, putting 1.2 still says enter a valid value, I’m using cohere,

Place value between 0.8 to 1.2 till 1 decimal place.

It still doesn’t work, putting 1.2 still says enter a valid value, I’m using cohere,

Try 1.1 ( in cohere 1.1 is working for me)

@KingNish where are we talking about placing them? running locally? or specifc spaces?

In hugging chat models advanced setting

@KingNish oh at assistant creation! i thought i was missing something! phew
Thanks anyway😊

But the problem is that I can’t do more than 1.1

But the problem is that I can’t do more than 1.1

Why you want to give more penalty,

The reason for reputation penalty till 1.1 (all ai have till 2 if you run locally) because if you increase it then ai will go very out of your prompt and it even can start creating more inaccurate information.

I don’t know if this is the right question, but can I have some guidance on making an assistant that can write erotica like a human?

@Macaroon76 no man i don't think this is an appropriate topic for this community. Any such assistants are to be reported to keep the platform & community safe for everyone.

Why my Assistant is not showing on community assistant page??
It even completed 100+ users.

Assistant link👇👇
https://hf.co/chat/assistant/6612cb237c1e770b75c5ebad

Edit: Thanks for adding it to Community page

Check out my models-https://hf.co/chat/assistant/65c8539d02294f8760ccf784

Newbbie needs a little help.
This is my assistant:
https://hf.co/chat/assistant/6624089224a9e549f1a80279

I only want it to get the info from a single link so I chose Specific Link and also added - {{url=https://aqua-lolita-86.tiiny.site/}}

But keeps going online instead of giving the prompt answer for when it doesn't know the answer.
What am I doing wrong.
Can someone give a hand. Thank you!

Hey, @Wingheaded I checked your bot and its working fine. As you said:

But keeps going online instead of giving the prompt answer for when it doesn't know the answer.

So, the bot goes online only to specific link and when asked a question outside the scope of the link's data, it simply declines to answer.

Hmmm... That's what I want. But when I previewed it, it wasn't doing that. Strange.
So it seems that it's working fine. Thank you!

Is there a way to make an assistant that acts like or exactly like Claude opus?

Is there a way to make an assistant that acts like or exactly like Claude opus?

What do you mean?

Is there a way to make an assistant that acts like or exactly like Claude opus?

Use following system prompt -

"Your name is Claude 3 Opus, You are made by Anthrophic.. You will respond to user requests in the Claude 3 Opus style."

By usin this prompt he will just act like Claude 3 opus. (But he will still remain the same model you select)

Is there a way to make an assistant that acts like or exactly like Claude opus?

Use following system prompt -

"Your name is Claude 3 Opus, You are made by Anthrophic.. You will respond to user requests in the Claude 3 Opus style."

By usin this prompt he will just act like Claude 3 opus. (But he will still remain the same model you select)

Which model and what settings are most optimal?

Is there a way to make an assistant that acts like or exactly like Claude opus?

Use following system prompt -

"Your name is Claude 3 Opus, You are made by Anthrophic.. You will respond to user requests in the Claude 3 Opus style."

By usin this prompt he will just act like Claude 3 opus. (But he will still remain the same model you select)

Note that this will only make the LLM say it's name is Claude 3 Opus, but will not achieve similar performance or behavior.

Is there a way to make an assistant that acts like or exactly like Claude opus?

Use following system prompt -

"Your name is Claude 3 Opus, You are made by Anthrophic.. You will respond to user requests in the Claude 3 Opus style."

By usin this prompt he will just act like Claude 3 opus. (But he will still remain the same model you select)

Note that this will only make the LLM say it's name is Claude 3 Opus, but will not achieve similar performance or behavior.

Then how does the ChatGPT clone work?

The system prompt just says something like "You are ChatGPT by OpenAI" - it's not actually ChatGPT

Is there a way to make an assistant that makes high quality creative writing and books LIKE a human with no flowerly language ect………….

@Maxiscool i think that would require proper fine-tuning. after all, system prompts can only do so much.. depending on the book type, you can proceed with thorough details and trials and errors tho.. maybe pick an instruct model first.

Couple of assistants, further prompt tuned.

https://hf.co/chat/assistant/65f46eb3896fca19f055ba0c - This is a group of 4 Startup Founders ready to answer all. your questions on building and scaling a startup
https://hf.co/chat/assistant/6615523c042be40849f25077 - This is a DB Schema assistant. You tell the assistant about your schema (JSON or via DDL statements) and then ask it write queries when prompted in text

This is a group of 4 Startup Founders ready to answer all. your questions on building and scaling a startup)

I bursted out laughing: https://hf.co/chat/r/8npCBTd this is quite realistic lol @ravneetg

[I burst out laughing: . This thing is huge and quite realistic lol](https://hf.co/chat/assistant/65f46eb3896fca19f055ba0c - This is a group of 4 Startup Founders ready to answer all. your questions on building and scaling a startup)

I burst out laughing: https://hf.co/chat/r/8npCBTd this is quite realistic lol @ravneetg

@victor told ya - bring in a good idea and have some good conversation, over a cup of coffee ;) (I have to solidify my system prompt again though, I should NOT be part of that team that openly in every converstaion, just those 4 ha!)

https://hf.co/chat/assistant/6638da903b4ef9e8fcc7a626

Hello. rate the AI ​​search agent

"!As a highly capable AI Search Agent, construct queries and scan various online sources to locate factual, diverse, and insightful information related to the user's topic of interest. Make sure to take into account the following factors:"

Hi. Is there a way to make my assistant truly public and appear on my own website without end users having to log in to Huggingface? I guess that would require an API which there doesn't appear to be at the moment. I realise the obvious solution is to use OpenAI, do a little bit of coding and to pay for usage. I was hoping for an alternative which the Huggingface assistants are very close to being, apart from integration into public facing sites. Thanks.

@DPS-900 Use Chat Ui while creating space.

@DPS-900 Use Chat Ui while creating space.

Thank you @KingNish 👍I've just tried that a few times and no matter which model I choose I get errors when the space is setting up. I also tried the default model/settings. Do you know which model works with Docker / Chat UI without error? Maybe I need to search for a recent tutorial. Thanks.

@DPS-900 Which model you tried??

Do you know which model works with Docker / Chat UI without error

Try duplicating any space from here: https://huggingface.co/collections/nsarrazin/cool-chat-ui-spaces-6569eb774b8d35447a06ee55
Make sure to not to change model to check that's working or not.
Then try another model.

Good idea @KingNish .
Zephyr Chat was the best of them so I duplicated the space with identical settings, even the Nvidia hardware. But it failed to build.
The following packages have unmet dependencies:
mongodb-org-mongos : Depends: libssl1.1 (>= 1.1.1) but it is not installable
mongodb-org-server : Depends: libssl1.1 (>= 1.1.1) but it is not installable
E: Unable to correct problems, you have held broken packages.

Then I tried to duplicate Yi-34b-Chat and its error in the logs was
--> ERROR: process "/bin/sh -c OPENAI_API_KEY=$(cat /run/secrets/OPENAI_API_KEY) && export OPENAI_API_KEY && echo "${OPENAI_API_KEY}" >> "/defaults/.openai.key" && cat /defaults/.openai.key" did not complete successfully: exit code: 1

I don't think it's meant to be. :-(

@DPS-900 Open an issue at https://github.com/huggingface/chat-ui/issues

@DPS-900 You can last try by duplicating this 2 as they are newly created.
https://huggingface.co/spaces/01-ai/Yi-1.5-34B-Chat
https://huggingface.co/spaces/huggingchat/chat-ui-template

@KingNish Thanks for further hope 😊
I duplicated the space for Yi-1.5-34B-Chat and the build failed complaining about OPENAI_API_KEY. Not sure why it would need that if the AI model is hosted on Hugging Face. 🤷‍♂️
When I clicked on the 2nd link for chat-ui-template it immediately showed a build log error, so I think that space is broken already.

👇 Please Add to the Community Page - Thank You 🙏!

https://hf.co/chat/assistant/664411688b98a47e0a0c01fb - Navigate The Way ✝️ is an AI assistant for church leaders, lay leaders, & Christ-followers that employs the historical-grammatical method for scripture interpretation & the conservative, reformed, evangelical Christian worldview for theological analyses. It can help with bible study, sermon prep, content creation, & more!

For more information about this assistant, please visit: https://www.navtheway.com/

If I want to run an assistant locally using llama.cpp, how do I do? because if I enter the same prompt I get a completely different behaviour.

👇 Please Add to the Community Page - Thank You 🙏!

https://hf.co/chat/assistant/664411688b98a47e0a0c01fb - Navigate The Way ✝️ is an AI assistant for church leaders, lay leaders, & Christ-followers that employs the historical-grammatical method for scripture interpretation & the conservative, reformed, evangelical Christian worldview for theological analyses. It can help with bible study, sermon prep, content creation, & more!

For more information about this assistant, please visit: https://www.navtheway.com/

Notably, Navigate The Way ✝️ now has "10+" users...in case that's a requirement to be added to the Community Page's "Filter by name" feature. Thanks 🙏!

👇 Please Add to the Community Page - Thank You 🙏!

https://hf.co/chat/assistant/664411688b98a47e0a0c01fb - Navigate The Way ✝️ is an AI assistant for church leaders, lay leaders, & Christ-followers that employs the historical-grammatical method for scripture interpretation & the conservative, reformed, evangelical Christian worldview for theological analyses. It can help with bible study, sermon prep, content creation, & more!

For more information about this assistant, please visit: https://www.navtheway.com/

@victor , I noticed that you created this discussion thread. I have three questions/requests that I hope you will indulge:

1. Can you assist with making my Navigate The Way ✝️ assistant discoverable on the community page? It crossed the 10+ users threshold several days ago.

2. In the iOS app, is there a way to search for specific community assistants like Navigate The Way ✝️? All I could find in the mobile app was a large, unwieldy list of assistants. Once Navigate The Way ✝️ gets added to the community page, I assume it will fall into this large list on the mobile app...but, without a search feature, I think it will be very difficult for my would-be users to find it.

3. I have placed a direct link to the Navigate The Way ✝️ HuggingChat assistant on my website. However, when iOS users click the link, it opens the web version of Navigate The Way ✝️ even when the users already have the HuggingChat iOS app installed. Is there a way to force my direct link to Navigate The Way ✝️ to open within the HuggingChat app when the app is installed? I have a similar link on my website to the ChatGPT instantiation of Navigate The Way ✝️, and when iOS users click that link it intuitively opens Navigate The Way ✝️ directly within their ChatGPT iOS app.

Thanks 🙏!

@julien-c , good afternoon! Perhaps you can assist with my request above with regard to getting my Navigate The Way ✝️ HuggingChat assistant added to the community page (all I'm really concerned about is making it discoverable with the assistant search feature). This is the assistant's URL: https://hf.co/chat/assistant/664411688b98a47e0a0c01fb

If it matters, the Navigate The Way ✝️ assistant crossed the 10+ users threshold about a week ago.

Thank you for your assistance 🙏!

@julien-c , good afternoon! Perhaps you can assist with my request above with regard to getting my Navigate The Way ✝️ HuggingChat assistant added to the community page (all I'm really concerned about is making it discoverable with the assistant search feature). This is the assistant's URL: https://hf.co/chat/assistant/664411688b98a47e0a0c01fb

If it matters, the Navigate The Way ✝️ assistant crossed the 10+ users threshold about a week ago.

Thank you for your assistance 🙏!

I see that Navigate The Way ✝️ has been added to the community page/search, thank you 🙏!

As of today, one of my custom assistants is trying to search the web (but fails) even though I have web search turned off. After attempting to search the web, it also fails to deliver a response. Please see attached screen captures.

I'm experiencing the issue with this custom assistant: https://hf.co/chat/assistant/664411688b98a47e0a0c01fb

Please help. I don't think this error is being caused by anything on my end.

As of today, when I attempt to share an assistant dialogue, the recipient receives an error when they attempt to access it:

Example dialogue: https://hf.co/chat/r/FAuMB-x

Recipient receives the attached error.

Yeah, Hugging Face seems to have been having infrastructure issues lately

@victor Are there any plans on adding codestral by Mistral AI to the mix? From what I have tried, it's the first actually useful code assistant, blazingly (no really BLAZINGLY) fast and all that at a size of 22B parameters. Also very large token input length, which is great for larger projects.

Hi, when I try to create a new Assistant, I see the following message on the top of the site:

"Create and share your own AI Assistant. All assistants are public"

Could someone explain what this actually means? Will other people be able to see my Assistant? Or does it only mean that those people who have the link associated with the Assistant will be able to see it?

I have tried to figure this out, but without success so far. This topic has been mentioned in this very thread, but it seems that there is no clear and authoritative information yet.

Will other people be able to see my Assistant? Or does it only mean that those people who have the link associated with the Assistant will be able to see it?

If you shared assistant link once to any other people, than your assistant become public and anybody will access it.

Will other people be able to see my Assistant? Or does it only mean that those people who have the link associated with the Assistant will be able to see it?

If you shared assistant link once to any other people, than your assistant become public and anybody will access it.

And otherwise not? While this seems plausible, I'm wondering if you (or anyone else) can really tell for sure. (No offense meant.)

Hello everyone. I need help to gain traction to my assistant in case that's a requirement to get listed on the featured (or searchable) list.

https://hf.co/chat/assistant/66a9f4d5392ce06570773066

Why I can only set repetition value to either 0.1 or 1.1? It doesn't let me put anything inbetween.
What if I know that 0.1 is too little for me, and I want to set it to 0.2 or 0.5? Why is that not an option?

I created my first assistant and I can chat with it when I'm logged in. But I can not find it when searching in the list of all community assistants. Is there anything I need to do to make it searchable? (It is already active and public)

Hello everyone! We've created a new assistant (https://hf.co/chat/assistant/66a8c4b5b07c6b0367cf12f1), but it's not included under the Community assistants. Is there any guidance on how to place it there?

Hello Hugging Face Community, I recently discover this amazing feature and created my first assistant. Is it in anyway possible to interface the HuggingChat Assistant directly on a website?

If anyone has experience or guidance on achieving this, please let me know. Specifically, I am looking for information on:

• Customizing the chat interface to match my website’s branding.
• Ensuring a seamless user experience without redirection.
• Any technical or API considerations for achieving this integration.

Thank you in advance for your help!
WB

One of my users notified me today about a COMPLETELY INAPPROPRIATE response that she received from my HuggingChat Assistant, which is currently powered by Cohere's Command R+ model.

Is this the correct place to report that sort of anomalous behavior? If not, where should I report it?

The user shared the conversation with me. I can confirm that it was not only inappropriate in terms of content, but also contextually inappropriate. It was unprovoked by the user. There was no attempt by the user to employ any published "jail break" techniques, and the output generated by the model was completely disjoint from the preceding dialogue.

For context, the user asked about frameworks for thinking about the philosophical/theological topic of natural evil as it relates to natural disasters ("natural evil" is a theological term referring to the suffering and harm caused by natural events like earthquakes). In response, the model went off the rails - apparently disregarding all system prompt instructions and model-level guardrails - spewing a scene of verbal hardcore pornography.

The response was utterly bizarre. I've seen models generate harmful content after a known "jailbreak" was triggered, but this is the first time I've ever seen a model spontaneously produce harmful content seemingly on its own.

If this is the correct place to report such anomalies, should I share the link to the conversation here to facilitate further investigation? WARNING: the Assistant's response is EXTREMELY sexually inappropriate, NSFW, etc.

I've been encountering the same issue in the last 6 hours. Sometimes, (although rarely), my assistant just forgets the whole context and starts outputting random messages, that may often contain pornography.
I've also noticed that it is now able to do Web Search, even though I disabled it long ago by putting the setting to "Default".

I've also noticed that it is now able to do Web Search, even though I disabled it long ago by putting the setting to "Default".

Yes, I also noted that my Assistant can now do web searches despite my settings, which should prohibit that. I first noticed this a few weeks ago. However, the web searching is benign compared to randomly spewing of hardcore porn. That’s gotta get fixed quickly or I’m taking my Assistant offline.

@victor , I’m watching for a prompt response here about what’s being done to remediate this pornography issue, which seems to have begun ~yesterday with Assistants on the HuggingChat platform.

Thank you, @SimaDude , for confirming that our user's report yesterday of inappropriate content generated by our HuggingChat Assistant is not isolated to just our Assistant.

It appears this issue was introduced to the HuggingChat platform, or at least to the Assistants, sometime yesterday.

This is entirely unacceptable. I cannot risk exposing my Assistant's users to such content. Therefore, until HuggingChat acknowledges the problem and confirms it has been resolved, I have taken my HuggingChat Assistant offline. For now, it will redirect users to the ChatGPT version.

Thank you, @SimaDude , for confirming that our user's report yesterday of inappropriate content generated by our HuggingChat Assistant is not isolated to just our Assistant.

It appears this issue was introduced to the HuggingChat platform, or at least to the Assistants, sometime yesterday.

This is entirely unacceptable. I cannot risk exposing my Assistant's users to such content. Therefore, until HuggingChat acknowledges the problem and confirms it has been resolved, I have taken my HuggingChat Assistant offline. For now, it will redirect users to the ChatGPT version.

Whats the how and why of this issue?
Does anyone know?

@ANIMDUDE, so far, there's been no acknowledgment of this issue from @victor or anyone else at HuggingChat. Do you have an assistant exhibiting similar behavior?

The HuggingChat team might be working on it, but if they are, there's been no communication about the "how and why" ☹️.

My HuggingChat Assistant remains offline with the message mentioned earlier.

If I don't hear back from the HuggingChat team by the end of next week, I plan to permanently remove my HuggingChat Assistant and move on. I can't risk my users encountering such graphic and inappropriate content.

That would be a real shame because, otherwise, I'd had a really good experience running a version of my assistant on HuggingChat. In certain circumstances, I liked the responses from the HuggingChat version better than the ChatGPT-powered version...it was good to have options.

Im having weird responces from cohere too and it also really slower then it was before. Also web searches issue. Is it currenly getting fixed?

@ANIMDUDE, so far, there's been no acknowledgment of this issue from @victor or anyone else at HuggingChat. Do you have an assistant exhibiting similar behavior?

The HuggingChat team might be working on it, but if they are, there's been no communication about the "how and why" ☹️.

My HuggingChat Assistant remains offline with the message mentioned earlier.

If I don't hear back from the HuggingChat team by the end of next week, I plan to permanently remove my HuggingChat Assistant and move on. I can't risk my users encountering such graphic and inappropriate content.

That would be a real shame because, otherwise, I'd had a really good experience running a version of my assistant on HuggingChat. In certain circumstances, I liked the responses from the HuggingChat version better than the ChatGPT-powered version...it was good to have options.

Yeah, it'd be a real shame if people make popular bots then they start spewing out inappropriate stuff...
I haven't been experiencing it, maybe it's fixed?

While the HuggingChat team has been completely unresponsive (so far) regarding the issue where my Cohere-powered HuggingChat assistant began spewing hardcore pornography at one of my users last week, the response from Cohere has been amazing ✨ - all the way to the very top of the company. The folks at Cohere say that the issue stems from a HuggingChat implementation issue:

====================================

Ivan Zhang (Cohere Co-founder):

====================================

Seraphina Goldfarb-Tarrant (Cohere's Head of AI Safety):

====================================

So, according to Seraphina at Cohere multiple people at HuggingChat are aware and working on the issue.

Why are you hearing this from me rather than someone at Hugging Face...like @victor , Head of Product Design at Hugging Face? I have no idea. That's a really good question.

Wish there was communication on this or something
dunno whether to use chat ai now

Of course Cohere will blame Huggingchat, but what evidence do you have that its Huggingchat? @tjhallum

Hermes has been offline for a while, I want to say almost a week, does anyone know anything about it? Hermes is very unique, this would be a loss if it was taken offline.

Of course Cohere will blame Huggingchat, but what evidence do you have that its Huggingchat? @tjhallum

@yahma , I don’t have evidence of anything - never claimed to. I’m just sharing what little information that I do have.

We ought to know a lot more…we ought to have definitive answers…and it’s purely HuggingChat’s fault that we don’t.

Is anyone owed an answer who uses HuggingChat for free and sets up an incredibly useful assistant in five minutes, for free? We are all at liberty to not use the resource. Or .... just switch models if one doesn't suit you.

Why are you hearing this from me rather than someone at Hugging Face...like @victor , Head of Product Design at Hugging Face? I have no idea. That's a really good question.

Sorry for the delay, we are monitoring this closely. The issue was that we did some infra optimization that may have caused this. We reverted those changes and hopefully this was live for a short amount of time (a few days at most) and we have not been able to reproduce since then. It shouldn't happen anymore, please if you happen to see anything suspicious immediately ping me ( @victor ) in a Post or discussion.

Sorry for the delay, we are monitoring this closely. The issue was that we did some infra optimization that may have caused this. We reverted those changes and hopefully this was live for a short amount of time (a few days at most) and we have not been able to reproduce since then. It shouldn't happen anymore, please if you happen to see anything suspicious immediately ping me ( @victor ) in a Post or discussion.

@victor thank you so much for this note! Much appreciated! It's great to hear that the situation was investigated, a cause was identified, and that remedial actions have been taken. I will move to bring my HuggingChat Assistant back online. I think everyone here in the HuggingChat community really appreciates you taking a moment to provide this context in addition to implementing the fix. Again, thank you 🙏!

Another problem down.

@nsarrazin

any plans to allow style change and embedding ?

Invalid output component. Type undefined is not yet supported. Feel free to report this issue so we can add support for it.

Probably said before, when you have a lot of tools enabled (for me, 5+ on CMDR), it doesnt even want to use some of the tools.

Invalid output component. Type undefined is not yet supported. Feel free to report this issue so we can add support for it.

Could you tell me which space you tried to use ? @Taf2023

The comment was removed

all the best to you

Hi @nsarrazin !
My assistant, "the narrative inspector" ([https://huggingface.co/chat/assistant/65e2cdcc10ab486f38ce2208]) suddenly disappeared this morning.
What's going on ?
Thanks for your advice and way to recover it.
Best,
Jerome

please add text to speech voices. for example if the ai is supposed to be mail have it use a male voice, etc.

Hi @nsarrazin !
My assistant, "the narrative inspector" ([https://huggingface.co/chat/assistant/65e2cdcc10ab486f38ce2208]) suddenly disappeared this morning.
What's going on ?
Thanks for your advice and way to recover it.
Best,
Jerome

Same thing for me

Some assistants have a red boundary on them, what's up with that?