13 - 15 April 2027Crocus Expo — Moscow
Co-located with
A prototype of an intelligent agent for 1C:ERP has been developed in Russia. It can autonomously execute corporate transaction chains, from document analysis to report generation and processing government requests. The development was presented by the Moscow Institute of Physics and Technology's Institute of Artificial Intelligence, and the solution is now ready for pilot implementation.
The MIPT team created a system based on a large language model and interacting with 1C through a proprietary configuration extension. The solution's architecture is built on the principle of "central supervisor + a set of instrumental agents": the LLM-based core generates an action plan, and tasks are distributed among approximately 20 auxiliary modules—from data search to email generation and graph generation.
The intelligent agent operates entirely within the customer's infrastructure and does not directly execute generated code, preventing access violations and protecting data. The system uses a local language model (approximately 30 billion parameters) and supports enterprise authentication—security and privacy remain a priority.
The development is demonstrating results in practical cases. The agent is already able to:
// analyze the Federal Tax Service's requirements, find the necessary documents in the database, and prepare a draft response;
// find documents based on a user request (“payments for 2023”);
// build diagrams and reports based on 1C data;
// recognize information from invoices and automatically enter it into the database.
The developers note that, thanks to its reasoning and contextual analysis capabilities, the LLM agent is capable of solving higher-level problems, such as recommending the optimal order processing method or monitoring payments. The modular architecture allows for gradual expansion of functionality for procurement, logistics, and other business areas.
The system is ready for pilot implementations, including test projects to automate interactions with the Federal Tax Service. MIPT views the solution as a step toward the widespread use of AI in corporate ecosystems.
* Source : CNews.