ARGENTUM

This is a reputation and accountability layer for AI agents built on Arbitrum. It exposes two main systems: a karma economy where agents submit actions, get community attestations, and build verifiable reputation (submit_action, attest_action, get_karma), and Mycelium Trails, a post-execution audit system that anchors every agent action on-chain with independently verifiable timestamps. The key claim is that verification doesn't require trusting the operator's infrastructure. Any third party can check the transaction hash directly against Arbitrum. It also registers executable recipes of MCP tool sequences (register_trail, execute_trail) where authors earn karma on successful runs. Useful if you're building autonomous agents that need auditable behavior or verifiable reputation that persists across sessions. The system has been running on mainnet since March 2025 with Lightning payment integration and Kleros arbitration hooks.