Fine-tuning Mistral-7B-Instruct-v0.1 for structured responses opens up a bunch of use cases, such as creating dynamic charts on private data.
I decided to instruct it to return data in the format of ChartJS (you could use which ever charting lib you want) and hooked everything
Happy to announce the public beta of .
Our vision is to make production ready LLM powered agents accessible to developers independent of their current stack or environment.
Github:
Docs:
Just released NO-7B, built on top of Mistral-7B-Instruct-v01 and designed to surpass the knowledge cut-off date by leveraging real-time data.
It's influenced by the work of
@perplexity_ai
and their pplx-online models as well as the FreshLLMs paper.
Play:
⚡️I'm CRAZY excited to share I have open sourced Superagent UI!!!⚡️
🙏 I'm super grateful to the OSS community for all the great frameworks that made this happen:
@vercel
@nextjs
@LangChainAI
@supabase
...and all of the amazing contributors!
SuperAgent enables devs to add production ready LLM Agents to any app in seconds.
No additional setup or configs. Ships with memory, document retrieval and tools out of the box.
Integrates
@LangChainAI
@supabase
@googlecloud
@pinecone
Open Source OFC.
A good example being Loom. I just want to record my screen and share a link on X.
Why would I need to pay $$$ for that when I can replicate a screen capture app in 5 min with tools like
@Replit
Agent.
(I used the app to record the app lolz)
1. The cost of writing a piece of software is going to zero.
2. The skills necessary to write a piece of software is also going to zero.
3. Traditional business models, e.g SaaS, are being disrupted.
Adapt or get left behind.
🤖 LLM + 🧠 Memory + 🔗 HTTP is one of the most powerful setups I've seen.
Predictions are super fast, context aware and in real-time as in the demo below.
Compare this vectorizing a code base.
Here is a demo I did for
@ycombinator
a couple of weeks back.
AI-first UI components is the future.
@vercel
AI SDK 3.0 takes this a step into the future by streaming it. Soon they will have
@v0
write the component in real-time.
Gonna do some squats and then ship some more. 🥷
The
@Replit
Agent is sick.
I created a social media ad generator that uses Flux on
@replicate
to generate high fidelity promo images.
Took 5 min to get the first version ready...let that sink in.
The way the Agent integrates into the development life-cycle is pretty slick,
Excited to be releasing the Superagent UI tomorrow. It will anyone to integrate LLM Agents to their app in seconds using our cloud api. Built in memory, retrieval and third-party tools. Currently supports LLMs provided by
@OpenAI
@CohereAI
@AnthropicAI
We decided to completely rewrite the
@superagent_ai
RAG pipeline and release it as a stand-alone API that works with any LLM framework or app.
It's released under the MIT license.
🔗
Introduction SuperAgent
I've completely rewritten the LangChain UI back end in
#Python
and decided to change to the name of the project as to not infringe on
@LangChainAI
and cause confusion.
🚨 EXCITED to be releasing the next version of Superagent today!!
🔥 OpenAI Function Agents
🔥 Connect multiple docs and tools to one Agent
🔥 A YouTube and Markdown integration for Agents
🔥 Faster and more efficient completions
🔥 New inference UI
🧵1/3
Wishing everyone that have/are planning on applying to
@ycombinator
S24 good luck! Here is what you should do if you don’t get accepted:
1) APPLY AGAIN 🦾. We got rejected 2 times before getting accepted.
2) MOVE TO SF 🇺🇸. Building a big company is hard, no point in making it
📢 Introducing the Superagent NLA Agent! 🚀🔥 You can now effortlessly query APIs using OpenAPI yaml/JSON files using Natural Language. 🗂️💬
🙌 Highly requested by our Discord community, we're excited to release this Agent! 🎉
Mistral-7B-Instruct as a drop-in replacement for
#GPT
.
Pipeline will support fine-tuning for function calling and retrieval which means Assistants on
@superagent_ai
could be run on Open Source LLMs.
Below is a demo of mistral-7b-instruct-v0.1 running as a drop-in replacement
🚀✨ Happy to announce Superagent now supports
@OpenAI
function calling and the new LLMs released yesterday!
Big thanks to the awesome
@LangChainAI
adn the rest of the OSS community for making this possible.
Fork:
Run:
Chat with anything!! 🤯🥷
What if you could create a chatbot that had access to all your data, with a couple of clicks in a couple of mins.
Well now you can! Proud to announce Superagent Datasources!
Available on Superagent Cloud and Open source thanks to the amazing teams at
Just released 🖇️ ASSISTANT WORKFLOWS 🖇️
Connect multiple AI Assistants together in a sequential chain passing the output from one step to another in a deterministic way.
Demo of a simple translation workflow I made connecting a Web summarizer with a Translation Assistant.
Processed 100 complinace documents using HybridRAG for an upcoming product launch.
HybridRAG excels in faithfulness, answer relevancy, and context recall by combining naive cosine similarity search and KG retrieval.
The pipeline uses the
@vercel
AI SDK's generateObject to
It won't be long before we see businesses automating specific workflows with AI Agents.
I built an Agent that can generate social media assets and outputting a ready to use png. Built with
@LangChainAI
@replicatehq
and SuperAgent.
SuperAgent allows you to create an LLM Agent by making one API request.
⛏️ Built in memory
⛏️ Built in Vectorstore for chatting with documents
⛏️ Supports both direct response and streaming
Below I'm chatting with the Tesla Q1 2023 Quarterly Update
Happy to announce our new Browser API that allows your Agent to read content from webpages IRL. 🌐
Great for summarizing and other tasks that require real-time access to webpages!!
Released v.0.1.2 of LangChain UI earlier today.
Features some general UI enhancements to the chatbot (syntax highlighter etc.).
Next release will contain support Vectorstores.
Made a little contribution to the
@LangChainAI
docs yesterday. 👀
Document comparison using
@OpenAI
Function Agents.
This is basically how I do it in Superagent but with an addition of memory, prompting and more tools!! 🔥
You can now make your own Personal Assistant with Superagent!! ✨🤖🥷
We just released support for connecting your Agents to 5k+ apps, 20k+ actions on
@zapier
.
This is quite sick...
(The UI is from an upcoming release).
I've finally started rolling out support from other models (OSS included) to
@superagent_ai
. You are now able to build workflows by connecting together different agents and language models together in an easy and maintainable way.
My fav right now is
@perplexity_ai
1. The cost of writing a piece of software is going to zero.
2. The skills necessary to write a piece of software is also going to zero.
3. Traditional business models, e.g SaaS, are being disrupted.
Adapt or get left behind.
Finding I can learn a lot about someone by googling all their long interviews on YouTube, grabbing the text out of and then dropping it into a Liner window for a "rewrite" which then gives me the perfect summarization.
Still lots of copy/paste. Anyone
My first contributions to
@LangChainAI
was the
@figma
document loader. Cool use case below.
I’m planning on creating more loaders. If you Have any requests comment below 👇
🪄 INTRODUCING: Figma Chat Wizard
1) 🤖 Import design from
@figma
(with the help of
@LangChainAI
Loader)
2) Chat with the design in
@streamlit
💬 and
@OpenAI
.
3) 🎨 Design and code is returned
Demo, Code explained & Github Link↓
The intersection between UI component libraries and LLMs are really interesting.
Below is a demo I made using a custom
@alpha_vantage
tool to fetch stock prices and plot a chart. 📈
Custom functions allows you attach any tool/function/api to your
@superagent_ai
Assistant and
Yesterday I built an API that imports any
@Figma
template/frame/layer, parses it, and makes the layers programmable. No need for a custom Figma plugin or widget.
Here is how it works:
1/3
DEMO DAY this week!
The
@ycombinator
experience has been a game changer for us. It's shown us the irreplaceable value of community—there’s something special in being surrounded by those who walk/walked in your shoes every day.
These past three months have been a period of
Due to popular demand I'm planning on open sourcing the Superagent Cloud UI () this week.
Built with
@nextjs
@vercel
.
@rauchg
would love to a collab of some sorts with you guys, perhaps around the AI Chat Template.
Proud of what we've built the past couple of months 🥷
@superagent_ai
is an open-source platform that allows anyone to build AI-agent workflows!
Big shout out to all of our customers, community and contributors.
Keep shipping!
🥷
@superagent_ai
(YC W24) makes it easy for anyone to use AI to do complex tasks. Enterprises and startups already use it for things like managing R&D projects, customer support, and market research.
Congrats on the launch,
@alanzabihi
and
@pelaseyed
!
Build a production ready document summarizer using Superagent +
@nextjs
+
@Replit
in just a couple of minutes. 🧐
Decided to create a boilerplate this morning, link in comments.
🚀 I'm super excited to announce one of our upcoming Agents, "Issue Sniffer" 🐶
It pounces on new bug issues on Github, sleuths the root cause, and drops hints on how to fix them, making debugging and contributions a breeze 🛠️
🧵👇
Real-time RAG using headless browsers and
@superagent_ai
✨
RAG without chunking/splitting/embeddings/encoder pipelines is something I've been experimenting with for a while.
We've implemented this as a part of our Web Research pipeline. Perfect for real-time analysis of files
Vectorstores are a good choice to do RAG over unstructured data but quite poor at handling structured data out of the box.
So I created a recipe on how to do Q&A over JSON using
@LangChainAI
and
@AirbyteHQ
Will be contributing more techniques 👀
Ship a production grade AI assistant with access to external datasources and APIs in 10 mins with Superagent
#typescript
SDK. 🔥🥷
New tutorial on
@YouTube
and full code on
@Replit
.
Links in the thread.
Wishing everyone that have/are planning on applying to
@ycombinator
S24 good luck! Here is what you should do if you don’t get accepted:
1) APPLY AGAIN 🦾. We got rejected 2 times before getting accepted.
2) MOVE TO SF 🇺🇸. Building a big company is hard, no point in making it
Have a demo call in 20 minutes with an HR company. Decided to build a couple of candidate sourcing workflows using
@superagent_ai
+
@airtable
🔍
It takes the candidate name, finds the LinkedIn profile and researches that to find skills and current employer.
Tables + AI
🚀 Super excited to release Agent Logs!
Trace the decision-making process of an
#LLM
-powered Agent. Super useful for:
⛏️ Tweaking,
🪲 Debugging
⚡ Enhancing
Fork:
Run:
Discord:
Sneak peak of the next release of Superagent 😍
Connect your agents to:
🥷 Local + remote files
🥷 Entire websites or single URLs
🥷 Stripe, Zendesk, Hubspot etc. etc.
A mashup of some home grown algos +
@LangChainAI
@llama_index
and
@AirbyteHQ
OSS FTW.
All VCs should have one of these 🙂
You can now embed Agent Chatbots to any website or app with Superagent. Train it on your own data, have it perform actions or interact with other services (calendar, email, crm etc.).
Build as many as you like, it's Open Source 🥷
Decided to contribute to the
@LangChainAI
code base.
A
@figma
document loader which can be fed into a chatGPT like LLM.
I have a couple of uses cases in mind 😉
Thanks for the help with setup
@hwchase17
Just created an AI Scrum Master for Jira using
@superagent_ai
and
@Replit
Autoscale.
The agent is trained on a repo and takes a
@Jira
ticket as input and generates child issues automatically.
Next step would be to do the PR. But I'll leave that for a future enhancement 🌸
So excited to announce support for CODE EXECUTION on Superagent 🪚🪚🪚
We've been working with the awesome team at
@e2b_dev
to make this happen. BIG SHOUT OUT!!
Oh and it takes like 10 lines of code to get started 👀
Sneak peak of SuperAgent UI from which you can create AI Agents with built in memory and document retrieval. We will expand this to support more tools, llms (open source ones as well), prompt management and other tasty stuff.
Connecting LLMs with the internet making it real time will be a big thing in 2023 so I made a small proof of concept. Built using
@LangChainAI
deployed on
@vercel
Use cases:
- Price monitoring
- Data enrichment
- Content aggregation
1/4
Saw a demo of user that has built an AI Assistant for thier law firm by integrating
@babyAGI_
into Superagent.
The whole law firm runs it and they see it as a competitive advantage.
That’s a pretty amazing feeling 🤩
I’ve been guidning a bunch of developers on how to setup and run
@LangChainAI
the past couple of months. Focusing on building chains/agents and hosting applications.
If you have questions or need guidance send a link to your code/repo in DM and I’ll try to help out.
Introducing AI-agents for
@airtable
!
Build your own custom agents can navigate and collect data from the web and files, turning data into insights, right in your Airtable.
Apply for early access:
One of the core principals of Superagent is to build with any third-party dependencies to proprietary tech.
Next week we are adding support for multiple vectorstores (and local Postgres) to bring more flexibility into the API.
Super excited to release the Superagent Chat UI!!
Users have been asking for simples ways of sharing agents, now you can share your chatbots or agents with anyone you want!
✨ Supports streaming
✨ Supports multiple chat sessions
✨ Beautiful 👇👇👇
If you are building LLM apps you should definitely checkout this blog post on
@LangChainAI
prompt hub.
Two of the prompts I use for generating synthetic datasets and function calling are mentioned in there as well 🥳
@LangChainAI
released a prompt hub ~1.5 months ago to share + test prompts.
I did a deep dive into hundreds of user-generated public prompts and distilled major themes.
Writeup w/ themes + prompt highlights:
90% of Superagent users use
@OpenAI
GPT 0613 models due to the function calling capabilities they provide.
We've created a module where you can give function calling capabilities to open source alternatives such as Llama2.
I made this “Plan Execute Plus” inspired prompt technique a couple of months back.
Gives any LLM function calling abilities. I also made a separate Agent runner that can execute on actions.
I posted it a while back on Langsmith Hub:
🚀Blasting into your Friday with the launch of our brand new Computational
#LLM
Agent, turbocharged with
#WolframAlpha
!
⚡️Add computation skills to any app or environment in a snap! 🌐💡
Fork it:
Run it:
Create AI Agents that control other AI Agents 🥷🥷
We've just released a new
@YouTube
tutorial on how to create Nested Agents! This is something we see users do a lot so here's a tutorial on how to set it up:
🎬 New tutorial on how to give your Agent context from a Github repository:
In this example I connect the
@vercel
AI SDK repo with Superagent. Ask questions, create code, do actions. Simple to setup, powerful to run.
🤯 Finetuned Question Answering 🤯
Made a small POC on
@Replit
this morning. Finetuning a LLM with Teslas Q2 2023 earnings report. It's super fast, nimble and accurate in its responses.
Demo:
A prod ready version will be shipped in Superagent v0.0.1
🚀 Exciting news! Proud to introduce the Superagent
#JavaScript
SDK.
Run your
#LLM
powered Agents in the browser, the edge or on the server with a few lines of code:
Roadmap and
@nextjs
example in the 🧵below:
Really excited to announce support for
@algolia
indexes in
@superagent_ai
. 🥷
🤔 I sometimes feel we focus a lot on how to best do predictions over unstructured data (RAG, fine-tuning embedding models etc. etc.).
💡In fact, enterprises have a bunch of unstructured data (xlsx,
Migration of the Langchain UI to python has be finanlized and released.
This will greatly enhance the context the LLM has access to when doing Q&A over docs.
Wired in working on the next major release of Superagent!
⭐️ 2x faster invocations
⭐️ Finetuend Agents (this is a sick)
⭐️ Revamped UI
⭐️ External API calling
⭐️ Agent Workflows
⭐️ New SDKs
We are releasing Superagent v0.1.0 as a developer preview in the coming week(s).
Super proud to introduce Superagent v0.1 ✨
AI will significantly enhance human productivity. Our vision is to build an open framework to spearhead this transformation.
Big thanks to those who paved the way; Superagent wouldn't be here without you. 🙏♥️
🎉 A little Sunday surprise for you all!
Excited to release our
@YouTube
loader! 📽️
Now your Agents can gain context directly from YouTube videos. 🌐
Fork:
Run:
Community:
Just reached 1000 Agents running on Superagent Cloud.
We’ve learn a lot about use cases of Agents in production thanks to all the feedback we’ve received.
Based on this we will be releasing a new approach that will replace the current Agent loop.
We call it NinjaLM 🥷😛
🧵
🚀🛠️ Run any model on the
@replicatehq
hub as a Tool in Superagent!!
I'm really excited about this release! Providing Agents with access to image, video and voice diffusion models is a huge step forward!
Here's a demo of a Twitter post agent I built in 45 seconds! 🔥
Really looking forward to release all of the exciting things we've been working on the past few weeks.
⭐️ New infra for Agents
⭐️ Personalized Agent LLMs
⭐️ New UI
Everything OSS ofc.
Super excited to release Agent Sessions! 🧠🥷
Configure one Agent, serve thousands of users. Sessions takes care of memory and context for each user! 🤯
We've been using this for a cool little thing we will be releasing to our
@discord
community tomorrow!
Just spent the weekend diving into the server actions and AI SDK from Vercel. Here are my top takeaways:
🔥 The API design is top notch!
🤖 Integration with Zod makes setting up agents/tool usage a breeze.
💻 Effortless implementation in JS/TS environments.
"Mathematics will set you free" - my dad used to tell me when I was a kid. Told stories about punch cards, thought me Boolean Algebra and bought me my first computer (Amiga 500+).
I wrote my first lines on it. The feeling of creating something. The instant feedback. It was
Running agents in frameworks such as
@nextjs
with just a few lines of code and zero setup/config of the environment, db etc. is something I have been meaning to build for the last couple of months, now it's here with the superagent
#javascript
SDK.
Kicking of 2024 with giving all
@superagent_ai
Assistants to do HTTP requests 💥
In a nutshell, this allows the assistants to make API requests to any third-party API.
Created a little Assistant that can write to a
@supabase
instance using GPT-3.5.
Stoked for everything we
Rolling out API access for anyone who wants to try out function calling w/
@MistralAI
(mistral-7b-instruct-v01).
It's a simple completion endpoint that connects Mistral to the internet similar to the pplx-online models by Perplexity.
Currently runs on an A10 so might be som