DeAgentAI

DeAgentAI is an on-chain infrastructure for creating and operating AI Agents with verifiable behaviour. It addresses three core requirements for agents in distributed systems: Consensus (a single authoritative output per interaction), Identity (one unique agent state at any point in time) and Continuity (reliable, traceable on-chain memory). The De(cision)Agent framework defines each agent by a Lobe (cognitive engine), Memory (genesis + evolving state) and Tools (capabilities and interfaces). Interactions are executed by Executors and finalised by Committers/validators to preserve identity and continuity.
How it works: creation publishes the agent definition on a host chain; users submit interaction requests; Executors run the agent and emit candidates; Committers select one canonical result and commit the state update; non-model logic is verified with ZK proofs, closed-model API integrity with TLS proofing/attestation, and open-model outputs via a hybrid approach with entropy-based selection; Memory includes short-term context plus retrieval from long-term history; built-in Tools include chain data queries, web access and a Decision Plugin for state-changing actions; approved decisions are executed via MPC; agents communicate using an Agent-to-Agent (A2A) protocol.

• Yves-Alexandre Kolter d’Ouradou (co-founder): leads technical and product direction for the agent protocol.
• Selwyn Zhou (co-founder): focuses on partnerships and strategy.
• Joe Z (co-founder/CTO): leads engineering across language models and agent systems.

• AlphaX: a flagship autonomous trading agent built within the framework that generates market signals and can operate strategies; it uses feedback to refine decision-making.
• Staking architecture:
  
  • Dynamic Rewards Hub: low-barrier, multi-activity rewards from a single stake.
  • Golden Shovel Pool: time-priority, tiered weight for long-term staking.
  • Community Arena: gated layer for higher-intensity, team or individual on-chain games tied to staking participation.