Just AI: Jay Flow Platform for working with neural networks based on multi-agent reasoning

The main articles are:

• Neural networks (neural networks)
• Generative artificial intelligence
• Speech Recognition (Technology, Market)
• Speech technology: On the path from recognition to understanding

2025: Launch of generative neural network platform

On February 21, 2025, Just AI announced the launch of Jay Flow, a generative neural network platform based on the multi-agent reasoning method. Jay Flow combines AI agents in the interface of an endless canvas and allows you to work in a single space with neural networks and content of various modalities.

𝓲

Фото: Just AI

According to the company, Jay Flow is a platform for "seamless" work with neural networks, based on a system of interacting AI agents. Unlike traditional models, where AI produces content at the request of the user, Jay Flow's agent approach allows neural networks to autonomously reason within a task, make decisions and use tools that not only process information, but also interact in a common environment. Jay Flow combines various neural networks and external functions, such as access to search or APIs of various systems through the environment of agents rendered as cards.

Individual LLMs, however powerful, are themselves limited in their interaction with the outside world. The breakthrough trends in GenAI for February 2025 are reasoning-based approaches and AI agents capable of planning, using external tools, interacting with each other and solving complex problems together. In creating Jay Flow, we tried to rethink the work of such systems from the point of view of the user, as well as understand how approaches to working with content are changing in the AI era. Inspired by Figma's idea of ​ ​ an endless FigJam canvas, we combined the content and data with which they work into a single environment of AI agents. told Dmitry Chechetkin, Director of Strategic Projects at Just AI

Agents are able to solve the problem by reasoning and using available functions, search the Internet, call specialized neural networks, create and directly execute program code. For example, if a separate program library is required to run the code, the agent is able to download it in the background, install it, and only then run the program. The Jay Flow interface makes it easy to connect AI agents to each other for their collaboration on multi-component tasks.

Among the neural networks available to agents are models such as, GPT 4o, Claude, Gemini, Flux image generators, Recraft, Ideogram, Kling, MiniMax and Luma video generators, ElevenLabs and Whisper speech synthesis and recognition models, and vision models for recognizing images and texts.

One of the main accents of Jay Flow is made on an intuitive interface. Jay Flow's workspace resembles a canvas in Miro or Figma, but is adapted to work with neural networks. Jay Flow supports a wide variety of content: texts, images, video, audio, tables, charts and code. Files can simply be dragged onto the canvas, linked to AI agents, and controlled the generation process. Content can be transformed, combined and adapted to different formats based on specific tasks and requirements. For example, turn an audio recording of a meeting into an infographic or mindmap diagram, a link to an article from the site into an audio podcast, an old family photo into a revived video.

The Jay Flow platform is available to both individual users and companies.