Inference Usage

og://network/inference

0inferences, run privately

Every request runs inside a hardware TEE enclave, relayed over OHTTP so even the network can't see who's asking. Inputs and outputs stay encrypted end-to-end — no one, not even OpenGradient, ever sees the data behind a prompt.

opg_spent: 0 OPG
inferences_served_today: 0
inferences_served_30d: 0
tokens_processed: 0

live tailstdout

How a confidential inference works

[01]

Sealed in transit

Each request is encrypted on the client and relayed through OHTTP, which strips identifying metadata. The network can route it, but never read it.

›ohttp.relay()

[02]

Run in a secure enclave

Data is decrypted and the model runs only inside a hardware Trusted Execution Environment. The host machine and operators are locked out.

›tee.execute()

[03]

Provably private

Remote attestation proves the enclave ran the exact, unmodified model. Inputs and outputs never leave the secure boundary in the clear.

›attest.verify()

Inference Requests

Daily confidential inference requests served through the network

276K

Jun 9→Jul 7

$og plot inference-requests --interval 1d --range 30dstdout

0└────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────

Jun 9Jul 7

Inferences per App

Total inference requests served per application

276Ktotal inferences