VK AI Space Launches Multi-Level Memory for Corporate AI Agents

VK AI Space introduced multi-level memory for corporate AI agents, allowing them to save, accumulate, and reuse information at all stages of work: both in the current dialogue and within long-term project work. This transforms them from static assistants into digital employees with sustained context and accumulated knowledge. The agent memory is embedded in the overall functionality of the VK AI Space platform. The multi-level memory consists of four layers: session memory preserves the dialogue flow and current tasks; user memory records the employee's profile and preferences; project memory accumulates materials, decisions, and rules within a project; and the agent's own memory reflects its personal experience—which approaches proved effective and what edits the employee made. This accelerates the preparation of complex responses, as context does not need to be gathered anew in each dialogue; the agent recalls similar cases from project memory. According to VK Tech estimates, the share of repeatedly solved tasks is reduced by 40-50%, depending on process maturity. For example, when preparing a tender, the agent generates responses to requirements from the knowledge base, and the employee refines and records them in project memory. Subsequently, ready-made artifacts and source data are reused by the team and agents, reducing time for information search and approvals. The memory does not open new access to information, which is important from a security standpoint: the agent only sees data available to the employee based on their role. Employees can view what the agent remembers and delete unnecessary data. All data is stored within the company's perimeter and does not go to external services. Roman Styatyugin, Director of AI Products at VK Tech, noted that agent memory turns employee knowledge into a scalable company asset, making professional experience reusable and creating a unified corporate AI space.