mantishack

agent runtime security - zero trust, zero setup, zero latency micro sandboxes

Local deterministic memory and safety kernel for AI agents. Replaces vector RAG with STRATA, a cryptographically signed, append-only ledger, so every agent state carries proof. Uses causal_walk engine to trace execution failures back to their root cause, and Operator Lite to block destructive terminal terminal commands before they ever run

Permanent memory for AI agents. Single binary, zero dependencies, MCP native.

Safe runtime for autonomous on-chain AI agents: isolated sandboxes, Library skills, encrypted secrets.