Localhost, but everywhere.

Simply, you pay for the routing and the ability to host with 1 button. The only case you would even need them in the first place is if you are a heavy user, or hosting it for a larger group of people. Prices are aimed to be affordable to anyone. Developer stands at 5$ per month and offers 7000 inbound requests per hour. You would need it if you want to have multiple API's and multiple machines hosting different stuff at the same time, to multiple people. The apis limit the request count NOT their size, which means that even if you are a heavy user you wont be hit with a usage limit unless you deliberately spam.

Ink host allows you to host every single model that is in the .gguf format. Coding, uncensored or creative doesnt matter. The app has a built in search tab for models and pulls straight from HuggingFace! Of course if you have your own tuned model in the .gguf format you can place it in the models folder and host it. It's one click anyways!

Long story short, if you have two machines, one strong enough to host an ai (you dont really need a very strong machines, i was left surprised by how efficient local LLMs have gotten.) and another machine that is just not strong enough as an example, a Desktop and a Laptop , you host the AI on the powerful desktop and enjoy using it from the laptop. You can use it from anywhere really, its stable, reliable and secured. As mentioned Ink only routes, it doesnt store.

I will be very detailed on this but first a few recommendations, when using public APIs from the Public tab in the site, always read the description and never trust it enough with any important information. As of if you are using a private API, it's secured, it's on your own machine and all conversations are stored there (encrypted using both host key and account key, meaning every user on that host has their history encrypted in a different way, this is made deliberately so the chat history cannot be unlocked unless both keys are provided. having the .db file itself is useless.), on your host machine. It never gets stored on our servers (Yes that means if you reinstall the PC you will lose all of your conversations unless you manually do a backup of the .db file and the .key file.). While chatting the messages are relayed through the servers (fully transparent on it, it means that ink acts as a tunnel from point A to B, its not E2E but it uses TLS protocols). We do not access the context of the message, they dont get cached, stored or reviewed in any way. There is no censor or filter on the messages (simply because we only route them). They are secured and can only be accessed via your account(more specifically, a key bound to your account). Outside users cannot receive the content of the message.

Mesh is a functionality that essentially allows the host to increase the capacity of their AI. To be clear if you host for many people, the load will get spread across all devices added on the mesh network. (Coming soon.)

The free tier is made for casual users that need the AI only for themselves or at most their friends, there is no need of mesh at such a low scale.

Not yet implemented. its planned for the credits to be like a currency allowing users to use paid API's hosted on the platform.

LS stands for Lost Sanches, just another very passioned project of mine. There is nothing else behind it. And the fact that it matches with the linux command is just a coincidence that i noticed recently.

The design is inspired mainly by github because it feel familiar to many people and I found it good looking and fitting the theme.

The site uses Python(FastAPI,Jinjga2) while the Relay is in Rust. Both apps Ink Host and Ink Client , use Tauri which is again Rust.

Compute points is a metric that can be set by the Host. When set it mesures how much resource the user uses and can apply stricter limits. It's available only on paid plans to allow users that want to host, more control over their network.

Currently only permanent limits that can be set are trough the API settings... limiting how many requests a single user can consume.