Solrouter is a privacy-preserving AI inference layer built on Solana. The protocol allows users to submit prompts to open-source language models without exposing the prompt content, model inputs, or agent state to any centralized provider. Queries are encrypted client-side in the browser using Arcium's RescueCipher. Encrypted payloads are processed inside a Trusted Execution Environment (TEE) running on Nosana's decentralized GPU network, where open-source models perform inference without the operator gaining access to plaintext data. Each query produces a cryptographic attestation that is anchored to Solana, providing a verifiable record that inference occurred inside the attested environment. The architecture removes three trust assumptions found in conventional AI inference services: prompts are not visible to a shared server, agent reasoning state is not held by a centralized provider, and settlement does not depend on trusting the inference operator. $ROUTER is the native token used to pay for inference on the network.