There’s new Cat in Town. Interview to Piero Savastano, Cheshire Cat, The Open Source AI Assistant Framework
Luca Lisci: What was your inspiration for creating the Cheshire Cat?
Piero Savastano: The idea for the Cheshire Cat was born while I was renovating my house. I was personally involved in the work, given my direct interest in the project. Around that time, ChatGPT was launched, and seeing the immense interest it sparked, I reflected on the potential of this technology. Looking at the building materials we were working with to renovate the house, I realized that what we had in front of us was only the base for building something much more sophisticated. This technology, called Language Model because it works with language, could be used, similarly to human capacities for attention and memory, for much more advanced applications.
I started talking about it on TikTok, explaining that what we were seeing was just the beginning. Noticing the difficulty in fully understanding these potentials, I thought of creating a clear and direct tutorial to help people explore and experiment with these technologies. Therefore, the Cheshire Cat was born as a tutorial, an attempt to spread the word to demonstrate that we are at the beginning of an era that could last decades, characterized by artificial intelligence. From there, growing interest transformed the Cheshire Cat from a simple tutorial to a semi-elaborate, open-source, and completely free tool, which, in the hands of a technician, can be turned into an artificial intelligence capable of performing specific tasks, such as ordering pizza, managing task lists, archiving store products, or assisting in learning topics of interest. The essence of the project, as a framework, is to create an artificial intelligence accessible to individuals, small businesses, and agencies, who wish to embark on specific initiatives without necessarily competing with digital giants, but still wanting to carve out their own space. I see artificial intelligence as a sort of evolution of websites: if before the innovation was in communication, now it moves towards automation.
Luca Lisci: What have been the main obstacles from a technological development standpoint?
Piero Savastano: The Cheshire Cat is built on tens of thousands of lines of code. The most complicated thing was to ensure that the Cheshire Cat was not strictly dependent on specific products. In its initial versions, for example, you had to use ChatGPT as a linguistic module. Then the Cheshire Cat had its own memory, its mechanism for selecting functions to use, its approach to conversation, its way, let’s say, of communicating with other applications. From a computer science perspective, there’s a lot inside, but it still depended a lot on ChatGPT as a linguistic module. Kind of like the Broca’s area in our brain, right? Our brain can do a bit of everything, and language is one of those things.
Despite the tendency in artificial intelligence marketing to present these models as if they were the entire AI, personally, I strongly disagree with this narrative. In fact, I think there is a certain deception in these statements. I’m aware that this happens, and in my opinion, it’s a distortion of the truth.
And the most complicated part was making the Cheshire Cat independent from specific solutions of certain providers. This means you are completely free to integrate ChatGPT, Google models, those from Anthropic, or any other, but you also have the possibility to set up a totally local system on your machines and use open-source models like LLama or Mistral. This aspect is particularly relevant when working with sensitive data. Imagine wanting to create an assistant to manage pharmaceutical, hospital, or legal data, which requires special protection. You certainly wouldn’t want this sensitive data to be spread all over the world. The Cheshire Cat offers you the possibility to keep your artificial intelligence completely internal to the organization and, being a framework, allows you to customize it according to your specific needs. For example, you can make it aware of the existence of certain drugs, their characteristics, or a particular drug archive that needs to be consulted.
Luca Lisci: So, the data issue seems very relevant to your strategy. What stance have you taken with the Cheshire Cat on ethical matters?
Piero Savastano: Data undoubtedly represents the most valuable component because everything else depends on it. When we interact with these new chatbots, what we send, technically defined as a “prompt,” is the sentence we send to this artificial intelligence. This prompt is processed by servers located on other continents, where the artificial intelligence resides, which then “responds” and sends back the answer. If, for example, we write questions about sensitive data in a prompt, this data travels and goes around the planet. Therefore, data is not only fundamental for building neural networks but also for formulating prompts for interaction. In this context, data assumes a primary economic value and runs the risk of being misused or traveling unintended routes. Therefore, my position is that everything should remain within one’s infrastructure as much as possible. As soon as technology and the cost of hardware, such as GPUs, allow, the goal of the community is to ensure that all artificial intelligence operates locally, without sending data outside its environment.
Luca Lisci: The Cheshire Cat is primarily a community. A motivated and participative community. Tell us a bit about it.
Piero Savastano: Our community is very diverse. The common element is definitely the tech enthusiast, people passionate about technology of all ages. We are essentially composed of technicians, ranging from the young twenty-year-old at the beginning of university, who finds participating in open projects more stimulating than studying textbooks, because it involves cutting-edge things, to seniors, people with also international experience. The interest is mainly related to technical aspects. Regarding evolution, the community is changing because the Cheshire Cat is becoming more stable and easier to use. As a result, interest is shifting from the construction of the tool itself to its actual applications. Some time ago, questions were mainly about how the project worked, what a linguistic model was, an agent, or concerned the technical aspects of the Cheshire Cat’s architecture. Now, a year later, questions are starting to be more focused on practical application: how can I put the Cheshire Cat online for twenty people to use? How can I make all this work on a server? The questions are more focused on how to make this vertical assistant accessible to non-technical people.
Luca Lisci: So, you foresee that future development will increasingly follow this direction, I guess. How do you see the Cheshire Cat? In which direction do you think it’s heading, or would you like it to go?
Piero Savastano: I would like the Cheshire Cat to follow the path of WordPress, which has shown over more than 15 years, that it can maintain a product accessible to a wide variety of users, from newspapers to kids wanting to start a cooking blog, through to educational sites and e-commerce. I envision the same for artificial intelligences: AIs that sell, inform, communicate, offer psychological support. I appreciate WordPress’ journey because they managed to keep the core of the project open, legally distinguishing the foundation from the commercial entity. This separation represents a strong message, and I would like to adopt this same strategy. They have been able to preserve the non-profit nature of the project, rewarding the community and maintaining good faith, ensuring that all the community’s efforts are directed towards a shared goal.
Luca Lisci: As the Cheshire Cat grows, how do you see the upcoming challenges?
Piero Savastano: I think one of the major difficulties will be keeping up with the digital giants. They are very fast and have their eyes on open source; anything that works well in this area is quickly incorporated into their offerings.
There is definitely competition with these giants, but the advantage is that they do not like the transparency and openness that characterize the more “socialist” approach towards artificial intelligence. We are in a period of strong centralization of technology. In my opinion, the biggest challenge will be adding visual capabilities to artificial intelligence, so not just limiting it to conversation but also integrating a sort of “occipital area” to the “brain”. Moreover, there is great enthusiasm in the community to explore aspects related to knowledge graphs and operational research. The world of neural networks is sub-symbolic, meaning the representations it creates of the world are distributed across layers of neurons, without specific entities for objects or people. On the other hand, in knowledge graphs and operational research, there are explicit symbols representing specific objects. An integration between neural networks and knowledge graphs would be a great victory, allowing us to manage conversations both in terms of nuanced language and shifting the discussion to unique world entities, like my dog Pallina or Rome Fiumicino airport.
Luca Lisci: Piero, tell us about the partnerships that have most significantly impacted the growth of the Cheshire Cat.
Piero Savastano: The partnership that was truly significant for us was with the German vector database called Qdrant, created by the same guys whose database was later used by Twitter for Grok. We collaborated with them because their database is open source, and we integrated it into the Cheshire Cat. When you download the Cheshire Cat, you find this database already included to store your data and conversations without having to spread them. This was our collaboration with an international hub. We have other partnerships in progress, but I prefer not to name them for now; it will be clearer when they are officially announced.
On the national front, we have participated in numerous conferences and meetups throughout Italy, talking about the Cheshire Cat. From an adoption standpoint, many small businesses have started using it to create products. The dream of enabling SMEs to create their products by relying on the Cheshire Cat, reducing development times from three months to a few weeks for a prototype, is being realized. There’s a certain movement in this direction, even if we don’t have an exact census, but the businesses that enrich their AI service portfolio with our support are in the tens, at the very least.
Luca Lisci: I have one final question for you. I’d like to know what advice you have for those, perhaps younger, who see in the Cheshire Cat and in your initiative a source of inspiration for their path.
Piero Savastano: My main advice is to tell the truth, not to be greedy, and to put culture ahead of everything else. These elements are essential. I believe that the key to success is doing something because you deeply understand its value, and this value does not necessarily have to be synonymous with money. When discussing with investors and middle management, often there are ulterior motives, but it’s understandable because we all need to make a living. If we want to seriously talk about value, then it must be linked to truth, to the freedom to allow people to accomplish what they intend to do without trying to confine them or exaggerate the potentials of what we offer, exploiting their good faith. For me, value means culture and, as you said, enabling others.
Luca Lisci: Before concluding, is there any point you would like to raise for our readers’ attention?
Piero Savastano: Yes, I would like to encourage our readers to reflect a bit more when they interact with technology, especially with this latest technological wave. Artificial intelligence, for example, tends to evoke psychological projections; we easily find ourselves thinking we are dealing with something almost alive, something new, because it is undoubtedly astonishing. However, my invitation is to try to look beyond the technology itself, to consider the people behind it and what their intentions are, and how we are involved in these interactions. So, not just to evaluate the technology for what it can do, but to understand that there are broad implications, not only technological but also economic, and, last but not least, issues related to work automation, social displacement. Who owns these technologies? Who controls the machines? Who can influence what these technologies communicate, including the types of bias they can convey? I don’t want to provide immediate answers, because that would be the subject of another lengthy discussion, but I want to encourage this kind of critical thinking.
Visit the Cheshire Cat website: https://cheshirecat.ai/