KinkzAI Control

—per hour

—per day

—per month

—vast credit

Loading fleet…

Rent a box

Under /hr Card memory 80GB and up

Cards any

Loading cards…

Host reliability 97% and up Download speed 200 Mbps and up Disk 64GB and up Include interruptible offers

Interruptible boxes are cheaper because someone can outbid you and your box stops — mid-conversation, for everyone on it. Fine for testing, risky for serving people.

Load right now

Checking every box takes a few seconds, so this only runs when you ask.

Boxes

Loading…

—coming in

−

—going out

—net

Loading income…

Other monthly costs

A dashboard, not your books. Export this into real accounting software at tax time.

Where it's going

Loading patron tiers…

Loading…

Accounts

Checking…

Say something

Banner Everyone sees it, every time, until you turn it off. For "restarting in 5 minutes" or "replies are slow right now". Notice Shown once per person, then gone for them. Runs for a set number of days so it also catches people who were away.

Show for days

Posted

Loading…

Publish now

Order Publish now

Posted

Loading…

What's connected

Checking…

How long setup takes

Checking…

Serving

Standard model

Served as Context Extra flags

Premium model

Served as Context Extra flags

Box image (decides what's already installed)

Nothing about the machine changes this — it's chosen when renting. vllm/vllm-openai:latest has vLLM but no nginx (setup installs it). Vast's own vllm template bundles more but is slower to boot.

Voices zip link Also install voice on model boxes

Off is right once you have a voice box. Voice on every model box reserves memory that comes straight out of KV cache — so it costs concurrent conversations on every card, not just setup time.

Takes effect on the next box you set up. Boxes already running keep what they were built with.