We are now available on
@huggingface
as one of the approved Local Apps which means now you can launch any models in Msty directly from within HuggingFace.
Please do us a favor and set Msty as your favorite Local App in your HF settings. We know it already is your favorite app :)
The next release will include this small but a very powerful toggle button. With just a click, you can make your Msty Local AI service available on the network, which means let it run on a powerful machine and interact from anywhere even from a mobile device.
A new, unique, and a giant feature is now being tested internally. No time to rest though. We are now tackling 2 big new features. The next release is going to be 🔥
Msty 0.9.0 is here bringing in many new features and improvements:
🤗 any
@huggingface
models
🦙 all
@ollama
models
🈁 reranking with
@JinaAI_
🔀 all models from
@OpenRouterAI
👐🏽 any OpenAI compatible providers including
@deepseek_ai
💄 Revamped Local AI UI
And a lot more!
Reranking is a technique used in RAG that significantly improves the results you get. Next version of Msty will make it very straightforward to rerank your results using
@JinaAI_
. Plug-in an API key and we'll take care of the rest for you.
Version 1.0 of Msty is now available! It brings in so many new features such as real time data, improved Knowledge Stack, New Attachments UI, new settings UI and oh so much more that even just the highlights turned out to be a full blog. 👇
While Meta prepares for a even bigger model, OpenAI casually drops a smaller but a powerful model.
#GPT4O
mini is available in Msty already!
✅ Text and Vision
✅ Solid Performance - reasoning, math, coding, multimodal reasoning
✅ Cheap
✅ Video, STT, & TTS coming soon! 🤯
Introducing GPT-4o mini! It’s our most intelligent and affordable small model, available today in the API. GPT-4o mini is significantly smarter and cheaper than GPT-3.5 Turbo.
In the next version, you'll be able to directly import Fabric prompts into
@msty_app
with just a click! It will download the prompts, fix the title, clean the content, and have it ready in just seconds.
Importing prompts from a folder will be possible as well.
#Gemma
2 is now available on Msty as well!
We make it very easy to use local as well as online LLMs on your machine.
✅ Offline First, Privacy focused
✅ Split chats with multiple LLMs
✅ RAG
And much more...👉
Gemma 2 is officially here! 🥳
Available in both 9B and 27B parameter sizes, Gemma 2 is higher performing than ever before. Download the weights on
@Kaggle
and
@HuggingFace
, or access the models in Google AI Studio.
Learn more →
Qwen has been one of our top 3 go to models and a model that we always recommend when being asked. It is really really good and Qwen 2 seems to be even better. In case you haven't noticed, it has been available in Msty as soon as it came out!
💗Hello Qwen2!
Happy to share the Qwen2 models to you all!
📖 BLOG:
🤗 HF collection:
🤖
💻 GitHub:
We have base and Instruct models of 5 sizes, Qwen2-0.5B, Qwen2-1.5B, Qwen2-7B,
“Msty is the simplest to configure and has most of the features anyone would want from an LLM app“ A quick overview of Msty by
@aicodeking
#ollama
#openai
#groq
We now have an official YouTube channel and our first video is about an upcoming exciting new feature in Msty. Go watch it here and please like and subscribe:
Msty Release day!!
The latest version comes with RAG feature - chat with your documents including Obsidian vaults🤯. Other features include
@perplexity_ai
API, search, concurrent chatting with local models using
@ollama
, split branching, and so much more!
The past few days, we noticed a sudden and huge spike in traffic from South Korea and couldn't figure out why... Turns out a popular Korean YouTuber has been making some amazing videos about Msty, driving all kinds of engagement our way!
In the next version of Msty, you'll be able to use any gguf models from
@huggingface
. And for your convenience, Msty will make the model compatible with
@ollama
. It would be as if you `ollama create` a model right within Msty.🤗🦙
Introducing Claude 3.5 Sonnet—our most intelligent model yet.
This is the first release in our 3.5 model family.
Sonnet now outperforms competitor models on key evaluations, at twice the speed of Claude 3 Opus and one-fifth the cost.
Try it for free:
Some great announcements from Apple
#WWDC24
yesterday. We took some notes on how we could also make Msty better with some inspirations from them. Here's us asking Msty to tell us about all the new announcements using the upcoming Real Time Data feature:
We took
@MistralAILabs
's new Cadestral model and compared it against GPT-4o to see how good it is at what it is made for - coding tasks. And we found it to be very powerful, accurate, and surprisingly, very concise. If you are interested in reading more and see some results👇🏽
And out of nowhere the latest version of Msty shows up. It brings support for head exploding speed from
@GroqInc
and head spinning latest models from
@AnthropicAI
and there are a few others for Ollama such as StarCoder2, dolphincoder and more!
Phi3 vision was just released - it is just 4.2b params and extremely impressive. 🤩
I feel this is a breakthrough for low-latency/live inference on image streams - just imagine what even smaller/more specialized versions of this will enable in robotics! 🤯
Msty will give you some insights about your RAG stacks. This is just the beginning and we plan to improve this further giving you even more insights about your RAG.
#comingSoon
#rag
#RAGDoneRight
After working on
@msty_app
for almost 6 months with a neck break speed, it was about time we paid some technical debt. We spent 2 days last week doing so and couldn’t be happier with the state of code now - less files, more consistency. And now back again on churning features 🧹
One of the users shared a screenshot of comparing the outputs from different models using Msty's split chat feature. I looked at the screenshot and wondered if the experience could be improved a bit. 👇
The latest version of Msty is now available. Prompt refinement, inline editing, branching, are just a few of the many features we have released in this version.
#ai
#chatgpt4
#mistral
#GenAI
So surprised and humbled to be on someone’s top of the list for an app that is just about a month old. We are into something! And we got more to come. 💪
“Msty fits the bill so well that we had to write a HOWTO post around how we got it installed on an M1 MacBook - and in that same post had to essentially disavow our previous suggestion to use Ollamac... 😆 Seriously nice work here with Msty.”
#localai
#ollama
#openai
#ChatGPT
Another small but very impactful feature - you will be able to split contents by either Characters (recursively + overlapping) or by Sentences.
We know
@Technovangelist
thinks that RAG is no good with character splitting so we wouldn't ship this without Sentence splitting 😅
@ptrykstr
Thanks. We had to clean up the list as it was getting longer and confusing. But we did add 1106 preview back about a week ago. Try updating local ai service which updates a list of available models as well.
@ptrykstr
For now, yes. We support more models than OpenAI so this makes it a bit complex and us just having a list make this simpler and manageable for us. Also, we will do our best to support new models as soon as they are out and this shouldn’t be an issue.
@findpurpose
Absolutely! What models do you use mostly? We can help you right now without waiting for us pushing an update. Mind hoping on our Discord to get this resolved right away?
@cibernicola_es
Ah! Just noticed the same handle. Got deceived with the profile pics :) This time we'll include you in the beta test for sure. Need to fix a couple of things first before the next round.
Announcing Codestral: our first-ever code model.
- Open-weights under the new Mistral AI Non-Production License
- New endpoint via La Plateforme:
- Try it now on Le Chat:
@JoakimThomsen
Sending those files as attachments to the model? This is only possible with some online LLM providers. It is on our list but not implemented yet. What models/providers are you using?
Read up on how Sujita, a fictional Online Course Instructor who is teaching three Power BI classes, would use Msty and
@MistralAI
to prepare for her course work without having to worry about cost, and most importantly, privacy.
@JoakimThomsen
Ah! Not right now. We have folders that you could add as a knowledge stack and after composing you can start chatting. One technical limitation is that it doesn’t have auto sync right now. We are looking into it though.
@stuli1989
Please let us know how much she likes it, how much of hallucinated info she is getting, and mostly if she has any feedback for us in how we could make improvements.
Soon folders in Msty will be much more powerful than just being a container. Why limit yourself to just grouping conversations when it could do more?
#powerup
🍄
@ClementDelangue
@Gradio
@argilla_io
You guys need a great app that you can call your own and something that is already integrated with HF 🤝. We just happen to know one. Just saying 🙃
@stuli1989
Hi! When there is an LLM involved, hallucination is always possible, unfortunately. You could minimize it by using a good model for inference, tweaking the prompt and some configs such as setting temperature to 0, etc.