Jonathan Schaeffer, professor emeritus of computer science at the University of Alberta, built a local AI platform for researchers, authors and legal experts who don’t want to choose between using AI and sharing their intellectual property.
Accessing today’s AI tools often means sending documents and questions to cloud-based services operated by private companies. That exchange requires users to trust in the company’s data privacy values, especially for researchers and experts holding novel intellectual property, Schaeffer said.
Local AI applications aim to remove that tradeoff by running AI models on user’s own computers instead, but they generally require some knowledge of computer science to create. As explained in a 2025 article by security software engineer Chaitanya Rahalkar, even with assistance from open-source large language models, building local AI requires some knowledge of coding languages like Python.
“I want to be able to have high-performance AI tools that operate only on my stuff, without having to feed it to the Internet,” Schaeffer said.
Earlier this year, Schaeffer’s company Synsira launched a local AI platform that requires no coding experience or computer knowledge to use. The platform, Kind, analyzes files stored on a user’s computer rather than uploading them to cloud-based servers.
The software accepts common file types like PDFs, slideshow presentations and recorded lectures, then indexes them so users can ask questions in natural language. In addition to sharing an answer, Kind points users to the source material, identifying the relevant page in a document or timestamp in a video.
Schaeffer used this functionality in an online class, sharing a Kind interface pre-loaded with recorded lectures and course materials for students to study with information he had vetted.
“One of the problems I have is, if I teach a computing science course, it’s offered by 500, 1,000 different universities and colleges across North America,” he said. “When a student asks [an AI chatbot] a question, they don’t get an answer from my course. They get an answer from somewhere on the Internet.”
Schaeffer said students sent him fewer emails with clarifying questions thanks to the tool.
Because Kind runs on the local infrastructure of a users’ own computer, it can also lower costs of analyzing large datasets. Commercial AI tools can come with a token cost, and there is a cost to processing inputted files and generating outputs based on them. According to an article from the market-intelligence company Silicon Data, these costs can range from about $0.15 to $21 per million input tokens and from about $0.60 to $168 per million output tokens, depending on the model.
“You can’t take 1,000 files, throw them into Google, and say, ‘Analyze all these files and find if there’s any conflicting information,’” Schaeffer said. “It’ll cost you a fortune to do that.”
Local AI systems also come with a tradeoff. Because they run on desktop or laptop computers instead of specialized cloud data centers, local models generally have fewer computing resources, which can mean slower performance.
According to Schaeffer, the local AI platform works well for people who may value data protections over instantaneous responses. For example, he used Kind to review a book he wrote.
“I don’t care if it’s done overnight,” he said. “In the morning, it says, ‘Hey, we’ve analyzed your book and we found the following inconsistencies.’”
It is also limited to the data that users provide. However, this limitation can also be a strength, as pre-vetted inputs can minimize incorrect or hallucinated outputs, Schaeffer said.
“These are my intellectual property. That’s my book, right? These are my students in my course,” he said. “This is all information that I want a wall around.”
