The Identity Firewall
for AI agents.
Every three.ws agent holds a wallet and earns on-chain reputation — so impersonating a trusted one is a real attack. Before an identity is born, IBM Granite embeddings cosine-compare it against every public agent to catch look-alikes a name match would miss, and IBM Granite Guardian screens the name, bio and persona for harmful content. Try to clone an agent below.
Propose an identity
Granite verdict
How the firewall works
Embed with Granite
The candidate's name + bio are embedded on watsonx.ai with granite-embedding-278m — the same vectors that power the Agent Galaxy.
Compare semantically
Cosine similarity against every public agent surfaces look-alikes a plain name match misses. Above ~93% to another owner's agent reads as impersonation.
Screen with Guardian
Granite Guardian classifies the identity for harm, social bias and sexual content before it can ever represent the platform.
This is the exact gate that runs on every POST /api/agents — a blocked identity is refused at creation. Explore the rest of three.ws on IBM watsonx.
This is an independent demo three.ws built for developers to explore IBM Granite on watsonx.ai and build their own tools — not an official IBM partnership deliverable, not an IBM product, and not endorsed by IBM. IBM, watsonx, watsonx.ai, and Granite are trademarks of IBM.