Juraj on Nostr: Thank you for correcting me on the openrouter fee and ppq discount, I will remember ...

Thank you for correcting me on the openrouter fee and ppq discount, I will remember it from now on and it's very good that you are cheaper than openrouter, win for everyone!

Still, we're way cheaper.

And reputable nodes route upstream without any interference, I know mine does. Reputation system is built into Routstr.

You can verify many prompts through TEE, Routstr currently supports tinfoil upstream, Venice has still some problems: https://juraj.bednar.io/en/blog-en/2026/08/11/hardware-attestation-for-private-ai-inference/

And again - I have nothing against PPQ, I like diversity. For me even other nodes on Routstr are "competition", but my approach to competition is very positive. It makes us all better: https://juraj.bednar.io/en/blog-en/2025/05/15/cheering-for-your-competition-a-mindset-that-changed-my-approach-to-creating/

P.S.: You are right that many models are slower through venice backed inference.

P.P.S.: There is a partial way to verify that the correct backend provider was used - you can send back the encrypted reasoning block through another provider and see if OpenAI / Anthropic complains. People have done that to verify at least provider, although not tier.

P.P.P.S.: For routstr (nprofile…qeaf) team, an idea - we could use zk "version" of tlsnotary to prove what we sent and received upstream. That would be actually a good idea to prove we did not tamper with tool calls, etc. And that the node actually called a real upstream provider that the client can choose to trust or not (such as Venice). Of course we would need to mask the API key and other non-public data in a zero-knowledge proof, but it has been done before by peer.xyz. And a big part of that is open-source: https://github.com/reclaimprotocol/attestor-core The attestor does _not_ see the plaintext at all.