Run the Pi agent on your own server: Pi pod keeps work sessions separate

Your own server. Pi already works with model providers connected by the user. Pi pod adds a server environment for teams: each session gets its own pod, and the owner sets shared configuration through a web dashboard. The author already uses Pi pod with his team and continues to develop it.

How to set up Pi pod

A standard installation requires a server with Linux, cgroup v2 and 8 GB RAM. The setup recipe also lists Docker Compose, git, openssl and Node version 22.19 or later.

Install the server with `git clone https://github.com/pi-pod/pipod.git`, `cd pipod` and `selfhost/upgrade`. The command `selfhost/add-user you@example.com --owner` creates the owner account.

On your workstation, install the client with `npm install -g @pipod/cli`. Then `pipod login --server <server URL>` connects it to the server, and running `pipod` from the project directory starts a sandbox and opens the session terminal.

After login, `pipod doctor` checks the server. The command `pipod credentials connect` connects model providers the user has already signed in to through local Pi.

How many sessions a server can run

Memory sets the limit. According to the documentation as of Oct 2, a standard pod gets 2 vCPU and 4 GiB RAM. A server with 8 GB RAM runs one such session at a time; a server with 16 GB runs three. For a server with 4 GB, the authors describe a smaller pod with 1 vCPU and 2 GiB RAM.

In the `/dashboard/` panel, the owner changes organization settings and policies. Members manage their own settings and environment templates.

The iOS and Android apps must be built from source using separate instructions. The server needs HTTPS to connect a phone.

Start updates on the server: `git pull && selfhost/upgrade`. Then update the client with `pipod update`, because the server rejects connections from a newer CLI.

Self-hosting is available now, while the developers' hosted service is still preparing to launch.