May 27, 2025
Release of LFAI & Data Voice Interoperability Initiative Open-Floor Inter-Agent Message Standard 1.0
The Linux Foundation AI & Data Open Voice Interoperability Initiative is happy to announce the publication of version 1.0 of the Open-Floor inter-agent message specification, along with two supporting specifications, Dialog Events 1.0.2 and Assistant Manifest 1.0. Together, these specifications provide a set of open, platform-independent messages that can be used by intelligent conversational agents to exchange information and collaborate on shared tasks through a simple, natural language-based API, following the metaphor where users and autonomous agents gather around a common ‘floor’ to converse with each other. At its simplest, the standard allows for a platform-independent interface between chat clients and autonomous conversational agents, but it goes way beyond that, supporting multi-user, multi-agent, collaboration across agent ecosystems.
While the current capabilities of agentic systems represent dramatic advances in AI technology, it is clear that there will inevitably be gaps in any one system’s abilities, both for technical reasons as well as for reasons of security, privacy, and boundaries in the ownership of customer relationships and data. This means that eventually, every agent, no matter how capable, will reach the limits of its expertise. Today, when this happens, users must interrupt their work to find a different and more suitable agent. The Open Floor protocols enable agents to publish their capabilities and query the capabilities of other agents. Newly discovered agents can be invited to a conversation, and the floor can be passed seamlessly between agents in different frameworks and ecosystems, depending on their expertise. Unlike most other multi-agent frameworks, agents do not have to have pre-configured relationships to interact with each other.
Version 1.0 of the Open-Floor specification finalizes the details of the inter-agent messages. A significant new feature is support for multi-agent systems, where several agents collaborate through the Conversational Floor, each supplying relevant information at the appropriate times, much like a human group conversation. Agents can be invited to participate in the conversation, volunteer information, respond to the statements by other agents, and can be excused when they’re no longer needed. This feature supports orchestration use cases where experts with different areas of expertise collaborate on tasks such as travel planning or patient care. For more technical information, please see our paper https://arxiv.org/pdf/2411.05828, which was presented at the 2025 AAAI workshop on Advancing LLM-Based Multi-Agent Collaboration.
The documentation for the 1.0 specifications can be found in our GitHub repository at https://github.com/open-voice-interoperability/openfloor-docs. We look forward to your comments.