the fact that confidential inference is in not the greatest of states around as seen in results at https://confidentialinference.net/attestation is one of the main reasons i'm building and sdk (still in early stage) https://github.com/enclava-labs/confidential-inference-sdk
the idea is to have a no nonsense way to integrate confidential inference properly, not just connecting to their openai compatible endpoint and praying for the best. this will enable you to fail close if attestation fails, it will hopefully also kick all providers asses to start doing things properly
the problem a lot of times is not necessarily that anyone is malicious, its just tedious to always have everything up2date and even more importantly we lack approaches that would be adopted by multiple providers so everyone rolls their setup a bit differently which leads to different attestation approach, different ways of communicating the versions etc.
also lets not get started about how easily reproducible the envs are, thats gonna be another story
aljaz
npub1al…hg9jp
2026-09-02 18:59:45 UTC
in reply to nevent1q…qle9
Author Public Key
npub1aljazgxlpnpfp7n5sunlk3dvfp72456x6nezjw4sd850q879rxqsthg9jpPublished at
2026-09-02 18:59:45 UTCEvent JSON
{
"id": "f0286260b8f443c985bc7e6b866dfc5d4a4e39e62f2ef276bfe750aefe4415d6",
"pubkey": "efe5d120df0cc290fa748727fb45ac487caad346d4f2293ab069e8f01fc51981",
"created_at": 1788375585,
"kind": 1,
"tags": [
[
"e",
"df9b6f85c984eb25100ec5346af9fa342c0a3fe575d06e37457351d450f514ee",
"",
"root"
],
[
"e",
"6c9d17c93c6d00732110a60a0450e8d25d62fb9fe27c154ece54523718b51c85",
"",
""
],
[
"e",
"e0fe0ae956c86156379e38f9e3db7287e081597a9df5c471baee05d53c34c593",
"",
"reply",
"576d23dc3db2056d208849462fee358cf9f0f3310a2c63cb6c267a4b9f5848f9"
],
[
"p",
"4e5622b575cdbb4d5ded093e48c68cd3f724fad547142f0c1b0aec2f2b2a0b2e",
"",
"mention"
],
[
"p",
"efe5d120df0cc290fa748727fb45ac487caad346d4f2293ab069e8f01fc51981",
"",
"mention"
],
[
"p",
"40b9c85fffeafc1cadf8c30a4e5c88660ff6e4971a0dc723d5ab674b5e61b451",
"",
"mention"
],
[
"p",
"52b4a076bcbbbdc3a1aefa3735816cf74993b1b8db202b01c883c58be7fad8bd",
"",
"mention"
],
[
"p",
"576d23dc3db2056d208849462fee358cf9f0f3310a2c63cb6c267a4b9f5848f9",
"",
"mention"
],
[
"p",
"4ad6fa2d16e2a9b576c863b4cf7404a70d4dc320c0c447d10ad6ff58993eacc8",
"",
"mention"
],
[
"client",
"nostria"
]
],
"content": "the fact that confidential inference is in not the greatest of states around as seen in results at https://confidentialinference.net/attestation is one of the main reasons i'm building and sdk (still in early stage) https://github.com/enclava-labs/confidential-inference-sdk\n\nthe idea is to have a no nonsense way to integrate confidential inference properly, not just connecting to their openai compatible endpoint and praying for the best. this will enable you to fail close if attestation fails, it will hopefully also kick all providers asses to start doing things properly \n\nthe problem a lot of times is not necessarily that anyone is malicious, its just tedious to always have everything up2date and even more importantly we lack approaches that would be adopted by multiple providers so everyone rolls their setup a bit differently which leads to different attestation approach, different ways of communicating the versions etc. \n\nalso lets not get started about how easily reproducible the envs are, thats gonna be another story",
"sig": "0ace1f4f14bba29f37ae363ee181a33aaf2abf7e518c21ae4be6feb5bee036aaf71c9e43343c6782157314db7de599d838a49cc481a0d80ba62ed09f6e0ba8f3"
}