The imecai multi-agent framework

All research conducted within imec.AI-labs is grounded in the use of a common platform that provides a strong, unified definition of what a multi-agent system (MAS) is [1]. By utilizing this shared codebase, the lab ensures that all projects align around core concepts—such as what constitutes an agent, how environments are formalized, and how interactions are structured [1]. This mutualized foundation reduces conceptual ambiguity, accelerates the development of prototypes, and strengthens the overall coherence of the lab’s scientific narrative [1, 2].

Principles of the platform

The imecai framework completely reimagines how decentralized agents communicate and operate. Its core principles include:

  • Communication via human modalities: Historically, multi-agent systems relied on highly structured, brittle data formats (like JSON or protocol buffers) requiring complex middleware [3, 4]. The imecai framework abandons this in favor of human modalities, using natural language and images as the unified, exclusive medium for inter-agent data exchange [5, 6].
  • Asynchronous messaging architecture: Rather than grounding agents in a simulated physical environment, the framework’s core communication mechanism is modeled directly after human messaging applications, similar to WhatsApp or Microsoft Teams [7]. This centralizes the rules for how information is continuously and asynchronously exchanged [8].
  • Decentralized, independent processes: Within this framework, an “agent” is defined simply as an independent, running process [8]. These autonomous software entities leverage the powerful reasoning and generative capabilities of Large Language Models (LLMs) and Vision Language Models (VLMs) to collaboratively solve complex problems [3, 5].
  • Native human-in-the-loop integration: Because the entire system communicates via natural language, the barrier for human participation is entirely removed [4]. An agent process can be fully autonomous or directly controlled by a human operator, allowing humans to seamlessly plug into the messaging architecture and collaborate natively with AI agents [8].

Upcoming release

The imecai framework will be released soon.