@woven pine if you followed the guide here then they should respond. August is a crazy month and there are many providers applying https://openrouter.ai/docs/use-cases/for-providers
#New Provider : Geodd
flint vector
woven pine
@flint vector thanks! we have checked all the checkboxes
woven pine
@violet spear Would you mind checking the status of our application? An update would be nice. Tomas from your team was the one handling before. Thank you
mild spade
Not trying to be rude, just trying to understand
How can users be truly sure that their data isn't being stored and is only used for the request and then deleted or that their prompts are private?
Is the backend open source so that people can review the implementation? How does Geodd make sure that they can't store prompts even if they wanted to?
woven pine
Thanks, I don't think we can integrate it with openrouter yet.
- Path /attest: Get a hardware attestation proving which runtime and model version are serving you.
- Blind routing: Your app encrypts the request body to your dedicated endpoint’s public key. Our gateway only sees routing headers, never the prompt. So even if the prompt is logged its not readable.
- We also provide third-party audit reports (e.g., SOC2 - still pending)
However I really like your idea of open-source code, this way the community knows the code and which version of git is running. Verify that the running binary is the same as what the compiler generates from the source code itself
mild spade
How does the attestation work code-wise?
Blind routing
Novel solution! Are models ran like a VPS? If you want to use a model, you have to launch your own private instance?
woven pine
Yes, exactly, they are in memory OS VM, so no storage. If we ever reboot everything is wiped
For attestation
Holds: a long-lived Ed25519 signing key KMS
Generates per call: a short-lived encryption key (RSA-OAEP) + session_id
Returns: a JWS (“attestation token”) that the client verifies with your public JWKS
We are working on bring your own HMS / KMS
mild spade
Yes, exactly, they are in memory OS VM, so no storage. If we ever reboot everyth...
Is it billed per token like Cerebras or is it billed like a proper VM? (example: $15 per 1 hour of running DeepSeek v3 0324)
woven pine
per million token
woven pine
Awesome, I like the feedback, keep em coming, I believe in "trust but verify"
I see what we can do with opensourcing the gateway, in terms of the VM its hard because we have custom Silicon GPU and not nvidia
mild spade
I see what we can do with opensourcing the gateway, in terms of the VM its hard ...
I assume the price per million tokens for modesl will be higher than most providers because of the overhead of manufactering custom silicon?
woven pine
Actually, no, it would be the same or less in most cases. Right now focus is not on making profit.
wheat spear
is there any way to sign up without passkey? I'm trying to make an account for the API
woven pine
Actually no password option for now, the idea is to store even the password in your device and not ours(hash). We just have the public key. If you have a phone with biometrics/face or fingerprint, you can use it. But good feedback
mild spade
What obscure/custom models do you want to deploy securely today?
No provider has Seed-OSS-36B. I'd like to play around with that model