Share a computer's models with your team

A computer with a large GPU, or a server in your office, can run models for everyone in your workspace. Install pasture on it, pair it, and share it.

Set up the computer

On a Linux server, connect over SSH and run the same command as on a Mac:

curl -fsSL https://packages.cinchapi.com/public/downloads/raw/names/pasture/versions/latest/install.sh | sh

On Linux, pasture starts when the server starts, and keeps running when nobody is logged in.

When Ollama is already installed as its own service, the command stops and says so, because that service holds the address pasture uses. Turn the service off, then run the command again:

sudo systemctl disable --now ollama

Install pasture shows how to keep the models that service downloaded.

Pair it

pasture pair --name "Office server"

The server has no browser, so open cinchapi.com/pair on your laptop or phone, sign in, and type the code that pasture prints. Use your models with Cinchapi shows the rest.

Share it with your workspace

In cinchapi.com, open your workspace's settings and choose Private models. Find the server, and share it with the whole workspace, or with chosen people. Only workspace owners and admins can share. Shared models cost no credits.

The people you share with see the server's models among their private models, with your name beside them. Cinchapi tells them that the owner can read the chats run on these models, because the chats run on the owner's computer.

Tools other than Cinchapi

To let other tools use the server's models over the internet, the server needs a public name, such as models.example.com, and port 443 open to the internet. pasture then serves HTTPS with API keys:

pasture https models.example.com
pasture key create alice
pasture install

pasture key create prints a key once. Give it to the person who will use it. Their tool connects to https://models.example.com/v1 with that key. Serve HTTPS on a public host has the details.