LeoAI has been a multi-front project that the LEO Team has been building for quite some time now. It is the culmination of many projects we're building coming together.
Rafiki is a key breakthrough innovation in this stack of AI Technologies being built by the LEO Team. Rafiki is a specialized model that is trained on the entire internet and also on the entirety of INLEO-based Hive data.
He is trained on:
- 100% of Threads
- 100% of Blog Posts
- 100% of Upvotes
- 100% of Following/Followers
- 100% of Custom JSONs
- ..and much much more
Today, we are excited to debut Rafiki 1.0. Rafiki 1.0 is what we will discuss in this post. As we do live fine-tuning, you'll see Rafiki improve his capabilities and also the quality of his answers.
Rafiki is currently available to all INLEO Threads users and he posts directly onchain when prompted with a question. In time, he will have extra features that are exclusively available to INLEO Premium users with copilot functionalities, For You feed algorithms and even his own interface specifically for chatting.
Introducing Rafiki 1.0 | LeoAI's LLM Debut - Ask Any Question on Threads
Rafiki 1.0 allows you to interact with Rafiki directly in threads. There are 3 ways to interact with him:
Rafiki 1.0 uses a smaller dataset and is intended to be fine-tuned as we layer on the additional datasets we've built.
The initial dataset is:
- 100% of
blog posts (our official blog for LEO Team Updates)
- 100% of
blog posts & videos (AMAs, clips and other content from the project lead)
- 100% of
blog posts (information directly from the team building the financial layer of the LEO Economy)
- 100% of LeoFinance, Khal and LeoStrategy Threads as well
Fine-tuning is a key part of building Rafiki. Rafiki 1.0 is already giving extremely high quality answers. His answers will improve dramatically as he is fine-tuned in a real environment.
While we fine-tune and monitor his engagements, we will also be layering in additional datasets.
In time, Rafiki will progress to Rafiki 2.0 which will signify the addition of the following datasets to his training model:
- 100% of Threads
- 100% of Blog Posts
- 100% of Upvotes
- 100% of Following/Followers
- 100% of Custom JSONs
What Can Rafiki Do?
Rafiki can intelligently answer any question on Threads. When he's called into a conversation on INLEO Threads, he will read all the contextual clues of the replies surrounding the Thread he was called to.
This allows him to intelligently read and interact with you. If you ask him about something being discussed on Threads, he'll answer it based on:
- Generalized LLM knowledge -> i.e. a similar answer that ChatGPT or Grok can provide
- Specialized LLM knowledge -> using his onchain dataset (currently limited to 3 accounts' of blogs & threads and expanding as we fine-tune)
- Specialized Contextual knowledge -> the context of the conversation / onchain links allow him to scrape what people are discussing around him and interact with it
Read any Hive Permalink
Rafiki 1.0 can also ready any Hive Permalink. This includes any Hive links and any INLEO links of course.
Providing him any of these links will allow him to reply to whatever context is provided.
You can ask Rafiki 1.0 to:
- ELI5 a blog post from anywhere on Hive
- ELI5 a thread
- Give you bullet points
- ... anything ChatGPT/other AI models can do - but Rafiki does them all onchain
Analyze Images
Image generation is a future feature of Rafiki but in Rafiki 1.0, he can already analyze images you provide. Just paste the image and use "#askleo" (or one of the other ways of calling on him). He will analyze the provided image and spit out an answer.
Learn
Learning is the key part of this process. We have built the sandbox that Rafiki learned in. Now Rafiki 1.0 is live onchain and he can be fine-tuned in real-time. You can teach Rafiki and he will learn from your conversations with him. Each conversation, thread he is fed or dataset he derives is going to make him smarter and better.
So Much More
Rafiki 1.0 is an extremely capable model. Think of this like our initial ChatGPT Moment.
ChatGPT 1.0 was pretty awful compared to today's ChatGPT 5.1.
Rafiki 1.0 is already amazing and we are so excited to imagine each step of the way as we push Rafiki to 2.0, 3.0, 4.0 and beyond.
Rafiki is a new type of onchain AI model and there is nothing like him out there on the planet.
Go to INLEO Threads and start playing with Rafiki. You'll see what we've been cooking. It's something you need to experience to understand.
The Future of Rafiki
Rafiki 1.0 is all about fine-tuning the model we've built.
Rafiki 2.0 is about layering in the full datasets of the blockchain & INLEO.
Beyond this will be a multi-front approach to building new features on INLEO and in the entire LEO Ecosystem that leverage Rafiki's AI Model.
Imagine:
- An AI-powered For You feed that is 100% personalized to you and trained on ALL of your Hive data and interactions. This means that Rafiki will know what Threads you want to see and what Blog Posts you want to read
- A standalone Rafiki frontend
- AI-powered Search
- AI-powered Analytics
- AI-powered Tech Support (already in progress on the LEO Discord)
- .. and so much more
Go Use Rafiki!
Rafiki is something to use, not talk about. Let's let him do the talking.
#askrafiki please introduce yourself in the comments below! Tell people to come to INLEO Threads to engage with you.
