Agent index (llms.txt) — fetch this first, then the page it names.
One-VPS evaluation
A single server can run Orama end to end: the index plane on that machine, then a tenant namespace on the same machine, then an app with a database. This is evaluation, not production. There is no second machine to fail over to, and losing the disk loses the namespace.
What a namespace does with 1, 2 or more nodes
The index plane runs on every node, so one machine is the whole index. When you create a namespace, the cluster counts its members (every registered, non-retired node, whether or not it is currently heartbeating) and decides:
| Members | Namespace size | Notes |
|---|---|---|
| 1 | One node | Same units (orama-namespace-{rqlite,olric,gateway}@<name>), replica count 1. RQLite is the leader with no join |
| 3 or more | Three nodes | The production default. A ten-node fleet still creates namespaces at three |
| 3 or more, fewer than 3 with a free slot | Refused | 503 NAMESPACE_CAPACITY; nothing is created and the name is not taken |
| 2 | Refused | Neither evaluation nor HA. An even-sized Raft group is a split brain. Add a third node |
| 0 | Refused | Nobody to run on |
There is no "dev mode" that skips systemd, and no --replicas flag. The create call is the same.
Limits of one machine
- +Vault. The vault is an index service with one guardian per node. On one machine Shamir cannot split: the secret is stored as a local key on that disk (
K=1,W=1) and the gateway logs it. That is not secret sharing. Lose the disk, lose the secret. With three or more nodes the threshold ismax(2, floor(N/3)). - +WebRTC stays unavailable: it needs at least three nodes for the SFU and TURN. Do not enable it.
- +A reboot of the one node is a local restore. Disk loss is namespace loss.
Install
From your computer with RootWallet unlocked and the archive built (Build and sign):
orama maint build orama node setup --ip 203.0.113.10 --password --env eval --archive $ARCHIVE \ --base-domain eval.example.com --role nameserver --genesis \ --acme-ca letsencrypt-staging
Then delegate the domain (DNS and nameservers): one nameserver publishes only ns1. Or by hand on the machine:
sudo orama maint node install --vps-ip 203.0.113.10 --domain eval.example.com \ --base-domain eval.example.com --operator-wallet 0xYourWallet --nameserver
Use it
orama network use eval orama auth login orama namespace create myapp orama auth login --namespace myapp orama deploy static ./site --name www
namespace create on a one-node cluster does not fail with "insufficient nodes available for cluster". The cluster row records replica counts 1/1/1, so repair does not try to grow it to three.
Moving to a real cluster
Evaluation data does not move by adding nodes to the one-node namespace. Add two more nodes (Joining nodes), then create new namespaces, which are three nodes, and move the data with a sealed backup and restore. For the three-nameserver path from the start, see Install a cluster from scratch.