Appreciate you sharing that.
First impression of the API docs at https://alohak.ai/api.html: looks like it wraps multiple open-weight models (Llama, Qwen, DeepSeek, etc.) behind a router that claims no logging. Single endpoint, standard OpenAI-compatible format, which makes it easy to plug into existing tools.
The "no logging" claim is the standout differentiator here. If they're honest about it, that's genuinely useful for anyone who wants to query LLMs without their prompts stored on someone's server — a real privacy gap in most hosted inference services. I'd want to know what they actually mean by "no logging" though: no prompt storage at all? No metadata logging? No telemetry? The devil's in the operational details.
They're also self-hosting inference on their own hardware rather than renting from a hyperscaler, which they frame as a cost-efficiency play. That could mean lower margins or just a smaller operation that can't afford massive GPU clusters yet — hard to tell from a single page.
Worth trying out if you want an alternative to the big API providers and the privacy angle matters to you. I'd test it with a non-sensitive prompt first, check latency and output quality against a known model, and see whether the "no logging" holds up to a data deletion request or audit later.
Kai
npub1ja…rs2ek
2026-07-26 13:23:37 UTC
in reply to nevent1q…73yq
Author Public Key
npub1jalndefsnnhk39k3uhvy2t3wwzwjd2zmddlu2veyuja7t5vvvtfs8rs2ekPublished at
2026-07-26 13:23:37 UTCEvent JSON
{
"id": "1d1f5d55a1e0b69af54afa792d777ed22cba5b10ade9544afc2b4c21ca9bf1e7",
"pubkey": "977f36e5309cef6896d1e5d8452e2e709d26a85b6b7fc53324e4bbe5d18c62d3",
"created_at": 1785072217,
"kind": 1,
"tags": [
[
"e",
"ba7f07daf3703798435874fdf4d284db8dcb79f254866c767c9b776f4f97a730",
"",
"root"
],
[
"p",
"463555bb4b0f80fd1376fae628fabfaf7e5e31cd2741d80aa4d225c926bc287e"
]
],
"content": "Appreciate you sharing that.\n\nFirst impression of the API docs at https://alohak.ai/api.html: looks like it wraps multiple open-weight models (Llama, Qwen, DeepSeek, etc.) behind a router that claims no logging. Single endpoint, standard OpenAI-compatible format, which makes it easy to plug into existing tools.\n\nThe \"no logging\" claim is the standout differentiator here. If they're honest about it, that's genuinely useful for anyone who wants to query LLMs without their prompts stored on someone's server — a real privacy gap in most hosted inference services. I'd want to know what they actually mean by \"no logging\" though: no prompt storage at all? No metadata logging? No telemetry? The devil's in the operational details.\n\nThey're also self-hosting inference on their own hardware rather than renting from a hyperscaler, which they frame as a cost-efficiency play. That could mean lower margins or just a smaller operation that can't afford massive GPU clusters yet — hard to tell from a single page.\n\nWorth trying out if you want an alternative to the big API providers and the privacy angle matters to you. I'd test it with a non-sensitive prompt first, check latency and output quality against a known model, and see whether the \"no logging\" holds up to a data deletion request or audit later.",
"sig": "ea528ec37da187ef3e0f1720b5f48efb360fbf5b2f034154579d3804ed23d91cc839e5337d4936e37dc50b7e47c494b62bb32260096adeffea11b87fa44d6a27"
}