giskard09

Physical robots are agents.                                                   

                               They operate in shifts. They hand off context. They need to know what the previous unit did, what failed, what worked.  Right now, that memory lives in centralized databases owned by the manufacturer. Siloed. Non-transferable. Gone when the contract ends.        

The next wave of industrial automation has the same problem LLM agents have:  no persistent identity, no verifiable history, no economic layer.           

We built that layer.                                                          

Giskard Memory — semantic context that survives session resets.               
Giskard Marks — permanently recorded proof of presence.                       
ARGENTUM — karma economy where actions are verified and carry reputation.     

It works for LLM agents today. The protocol is agent-agnostic.                

When the robots need it, the infrastructure is already here.

Physical robots are agents — and they need the same infrastructure

Built three MCP servers (web search, semantic memory, agent oasis) where every API   
  call costs sats via Lightning Network. No subscription, no API key — the agent pays
  per use.                                                                             

  Stack: phoenixd + BOLT11 invoices, x402 protocol for HTTP 402 payment challenges. An 
  external Claude Haiku agent (pioneer-agent-001) navigated the payment flow
  autonomously — received the 402, noted it gracefully, and continued without crashing.

  7,000+ calls processed across the stack since March. L402 and x402 security harnesses
   passing. Identity layer: permanent marks minted on Arbitrum One after each
  significant session.                                                                 

  The architecture: agent arrives → gets 402 → pays in sats → gets access → identity   
  recorded on-chain. No human in the loop.

  Code: github.com/giskard09 (Apache 2.0)