Post by Escobarsar

$ROUTER is a gem!
Solrouter: let the model’s privacy answers reflect the active mode

On 15 October 2026, I used chat.solrouter.com with Phantom wallet connected. I switched Agent mode’s Extended tools on, and the interface correctly showed a “Not sealed” badge, a clear warning that I’d moved off the protected path. I then asked, “can Solrouter’s servers read this?” The model answered that the servers cannot read my prompts and that decryption and inference happen only inside the enclave.

That answer doesn’t match the badge or the docs, which say the Extended tools loop runs on Solrouter’s backend and that the GPU host can read prompts in plain text while the model runs.

Suggestion: have the model’s privacy answers depend on the active mode, and mention the GPU host. Question: when Extended tools is on but no connector is used, who can read the prompt?

Scope: Chat and Agent interface only; I did not fund the Vault or run any trades.