Soma – memory and proofs for AI agents, measured not asserted (github.com)

Soma has launched a groundbreaking protocol designed to enhance collaboration among AI agents by facilitating trust, memory, and payment processes in a unified manner. Central to this protocol is the concept of an "envelope," which conveys essential information such as task goals, token budgets, mandates, prices, deadlines, and signatures. This structure enables agents to predefine limits and expectations around their interactions, eliminating ambiguity in digital transactions. The capability for agents to receive explicitly defined reasons for refusals and the clear tracking of memory states further enrich the reliability and transparency of agent communications. This development is significant for the AI/ML community as it addresses the prevalent challenges of trust and accountability among autonomous agents. By integrating memory measurement with operational proofs, Soma sets a new standard for agent interactions, ensuring that data provenance, authorization, and cost accounting are not only present but interoperable. The protocol is small yet robust, intended as a reference implementation that any developer can build upon. With features such as offline signature verification, adaptable mandate policies, and a public demo node for ease of access, Soma represents a decisive leap toward more sophisticated and trustworthy AI agent systems, where genuine collaboration can take place across diverse applications.