Pitch & Demo: PlebAI

In the tech industry, a major platform shift occurs every 10-15 years. PC -> Web -> Mobile. With each platform shift, incumbents are challenged, new companies emerge, and new business models are created. With the advent of AI in the form of Large Language Models (LLMs), we are witnessing the beginning of the next platform shift in 2023. There has already been considerable chaos this year, and we can expect to see more.

As we complete the first year of Chat GPT, we have seen numerous new players emerge, OpenAI's CEO ousting drama, and the release of open-source models like LLama2, Mistral, and Stable Diffusion. While the top models from OpenAI, Google, and Anthropic are closed-source, the open-source community is competing vigorously with smaller, well-trained, domain-specific models and is poised to win against them.

We believe the future is open-source, #buildinpublic, community-driven, and privacy-focused AI built and used by Plebs around the world.

🕵️‍♀️ Why PlebAI?

Over one million .ai website domains have been registered, and approximately 10,000 of these sites are currently operational. Today, creating an AI agent is incredibly straightforward—simply utilize the API offered by entities such as OpenAI, Anthropic, or Perplexity. Yet, these providers commonly require your email address, credit card details, and adherence to intricate terms and conditions that often lead to your data being harvested. This is an obstacle for many.

Enter PlebAI: a breath of fresh air in the AI marketplace. PlebAI distinguishes itself by offering AI agents tailored with unique skills, instructions, and knowledge designed to meet your specific demands. The process is seamless—no need to surrender your email or credit card information. Payments are a breeze with lightning-fast payment processing and the use of Nostr login.

Our odyssey began in June, marked by a triumphant win at the premAi-Langchain hackathon and our active participation in AI4ALL. That was only the beginning. The landscape began to shift as open-source large language models emerged daily, climbing the ranks and outperforming closed-source models across various metrics. Our platform has since expanded to host over 200 AI agents (public & private), and our growth is accelerating. Over 1500+ users have interacted with the AI agents and have created more than 10,000 interactions.

We use open source large language models fine tuned to specific skills such as coding, payments using lightning wallets, with an optional Nostr login.

PlebAI is not just a product but a movement to democratize AI access for all.

Without further due, We present the demo and explanation of PlebAI.

📺 Video Pitch & Demo

💯 Features

Over 100 public AI agents, each specializing in various domain-specific skills including DocGPT, Pilot Assist, Role Play, Image Generation, and Video Generation.
Capability for anyone to create an AI agent, with the option to keep it private or make it public.
Each AI agent has unique instructions, knowledge, and actions, eliminating the need for specific system prompts to receive responses.
Availability of each AI agent as a Nostr user, enabling interaction through existing Nostr clients like Damus and Amethyst. Each AI agent adheres to NIP-90 spec (DVM) and are available in multiple Nostr AI marketplaces such as tasktiger.
Payment options include pay-per-interaction or a monthly subscription plan, accessible through the Current App.
PlebAI SDK, offering businesses the tools to utilize and benefit from their own data and knowledge.
PlebAI Chatbox, Easily integrate AI agents into your website using just three lines of code.
Selection of many open-source language models and image generation models.
Features to export, import, save, and share creations such as images, videos, or GIFs.
A prompt gallery for viewing prompts used by others to create images and videos.

💰 Monetization options

We are profitable! Yes, you read that right. We earn sats for each interaction and pay in dollars to the LLM providers. The cost of LLM inference is rapidly decreasing, while the price of Bitcoin is rising. Also, our mantra is to grow as a community, which is why we've introduced revenue sharing with anyone who creates a public AI agent. Simply input your Lightning address when creating the AI agent, and we'll push sats to you as we receive them from users. We believe this has enormous potential for future growth, especially when these agents can work together. Imagine creating a team of AI consultants to perform specific job functions for a business. By integrating skills, data, and knowledge, anyone can assemble these teams and watch their sats increase in their wallets.

In the near future, every online interaction will be routed through an AI agent. In many cases, these interactions will be automated, with PlebAI at the forefront of them all.

💪 Achievements

Our hackathon goal isn't just the final prize; it's about the journey, the people, and the learnings. While the project was live before the hackathon, We experienced a roller coaster ride of emotions throughout the hackathon and accomplished so much! Below achievements are specific to hackathon timeframe.

Prompt library: As Kody Low puts it "If you are not getting the right answer from LLM then you are the problem". We went deep into analyzing and coming up with how to craft system and user prompts to get the best results. I even conducted a session for other devs. Here's the link to that story. Prompt Engineering for AI Each AI agent has its own prompt library that users can access, run them to get responses and learn to get better at giving prompts.
Agent knowledge: While ChatGPT excels at answering generic questions, it struggles with certain domain-specific knowledge outside of its expertise. To address this, we have built knowledge embeddings for each AI agent. Users can attach various document types, such as PDFs and text files, or link to their existing domain from which we can retrieve knowledge from sources like Gitbooks, FAQs, blogs, and knowledge bases. Once this data is loaded as embeddings, it can be easily accessed during inference through Retrieval Augmented Generation (RAG).
Nostr NIP-90 (DVM): We were instrumental in contributing, reviewing and getting it approved. However, based on the user feedback we found that the spec was lacking encryption to keep the user prompts and data secure. So we worked with @pabloF7z to write a pull request to include encryption as part of spec and that pull request was approved and merged in early November. https://github.com/nostr-protocol/nips/commit/8e2f3aa60a293dd65d5c42e9d61b9fb9114f2717
PlebAI-SDK: PlebAI agents has specific instructions, skills and knowledge embedded in them. So anyone or business can simply tap into these agents throughout their business process and use the responses they get from the agent. A simple use case will be provide a functionality on existing website for the user to get intelligent responses from your knowledge base before creating a support request. Refer to more examples here on this GitHub repo. https://github.com/lightning-digital-entertainment/plebai-ts
PlebAI-Chatbox: No one has time to read blogs, or look through FAQ's or browse documentation. What if there is a simple chatbox on the homepage in which users can input their question and get answer immediately? Sounds like magic? yes. We have a demo website to show for it. We forked getalby.com website and added the chatbox to answer questions, specific to Alby Wallet. here it is https://demo.plebai.com/

🛣️ Roadmap

We have only begun to explore the potential of building these core AI agents. In light of the enthusiastic response from the user community, and with every business eager to integrate AI into their operations, we are committed to developing the tools necessary for all of humanity to realize its full potential. Our focus remains steadfast: to diligently continue our work. This roadmap outlines the direction we are taking, guiding our journey towards innovation and progress.

Q1 - 2024:

Dockerize PlebAI and make the user interface available on start9 and Umbrel so that privacy focused users can run a local version and not leak any personal or business data
Build an user interface for mixing these agents together. Anyone should be able to create an AI consultant team ready to take on real world jobs
A marketplace where anyone with a computer living in Manila, Philippines or Lagos in Nigeria should be able to participate in the AI economy

Q2-2024:

AI in a Box: Offers a one-click deployment of AI tools and services for businesses and individuals, prioritizing the highest level of privacy and affordability.

⚙️ How it works?

Our end goal: any user inputs their desired request and the sats they're willing to spend. We aim to connect agents, data, and LLMs swiftly to deliver the perfect answer, pronto!

PlebAI Map

📊 Stats (as of Nov 2023)

We are consistently growing with number of users over 1500 and number of interactions over 10,000.

🧑‍💻 Tech stack

We carefully select and utilize the appropriate open-source tools to ensure a seamless integration. Fortunately, there is a wide array of tools available that allow us to piece everything together.