rivetDedicated server
A Debian 12 or Ubuntu LTS machine with systemd and sudo access, on an address your nodes can reach.
Server setup
A hub that’s always on.
Put RivetHub on a server, VM, or Linux container so your memory is always on and shared by every machine you work on. Two installers do the work: the datahub first, then one node per machine.
Setting up your own laptop? The Quickstart is the shorter path.
Step 01 · the host
Any Linux box with systemd.
Virtual machine
The same distributions inside a VM. Give it a stable address; the nodes will enroll into it over SSH.
Container
A Linux system container with systemd. On a Docker host, the installers’ --docker option runs services in containers managed by the host’s systemd.
Disk, networking, and dependency details are in the datahub guide and node guide.
Step 02 · the datahub
Install the datahub.
The datahub is the center of your setup: Postgres with pgvector, the memory schema, and the certificate authority your nodes trust. No agents run here. Its installer runs from a checkout of the distribution repository:
$ sudo bash install/datahub.shThe wizard asks three things: who owns the hub, which memory mode (lite needs no inference services; full adds embedding and compaction endpoints you choose), and bare-metal or Docker Postgres.
Step 03 · agent nodes
Enroll your first node.
The node runs the RivetOS runtime — this is where your agents live. It enrolls into the datahub over SSH and receives its certificates in one pass. On the node host:
$ curl -fsSL https://get.rivethub.io/node.sh | sudo bash -s -- --hub owner@192.0.2.10Replace owner@192.0.2.10 with an SSH login on the datahub that can run rivethub-hub. Repeat on each machine you want agents on.
Step 04 · verify
Check it, then get to work.
- On the datahub:
rivethub-hub status. - On the node:
systemctl status rivetos-agent.service, thensudo -u rivet -H rivetos doctor. - Connect an app, run a small task in a connected tool, and watch the session appear in your hub.