Intelligence ≠ authority
The AI can think.
AgentBank decides what runs.
A secure execution layer for agents: your local side stays untrusted by design (no keys, no OAuth). Natural language → intent → policy → Auth0 Token Vault & Base Sepolia — only sanitized results return.
Not a chatbot — an identity and permission system for AI agents.
Local agent — untrusted by design
Represents real-world local AI: intentionally restricted — no API keys, no tokens, no direct access. Requests go to AgentBank; credentials never live here.
AgentBank — authority & execution
Policy allow/deny/step-up, then execution only if approved. OpenAI routes language to intent — it does not act. Token Vault + chain handle the real work.
Auth0 Token Vault
Google and GitHub OAuth tokens are stored by Auth0 — never in application code. The intermediary retrieves them on demand. OpenClaw never sees a raw credential.
Consent Delegation Log
Every Token Vault call is logged with timestamp, service, and data summary. Step-up auth gates large financial operations. Humans stay in full control.