Privacy Issues by Using Cloud-Based AI Agents. Modern cloud systems offering personal text/voice AI agents cannot guarantee the confidentiality of your personal data. We have many examples from ChatGPT, Grok (AI) etc.

Is building a local AI agent the solution? Yes and no: To avoid confidential information leaks, we can create a standalone secure private or corporate AI agent.

Modern open source alternatives already allow us to implement all AI agent functions locally, and, most importantly, we don’t need high-performance hardware.

Unfortunately, the proposed scheme still leaves open the question of licensing certain elements of the system for commercial use, and a solution of this problem is not covered in this talk.

In this talk we goes through all element of local text/voice AI agents:

• Wake word system activator
• STT, OCR recognizer, Message collectors as input channels
• Small LLM model as an command Input corrector
• Agents Schort time and long time Memory
• Large LLM model as part of the decision-making system
• Outputs channels like a TTS, Messaging System etc..

In additional can be discussed:

• Models Fine tuning part
• Static commands as a predictable parts of decision-making system.

After the talk, the participants will have a ready-made local AI agent with a description of the obstacles that currently prevent the mass distribution of local offline agents.