Verify it works

Two commands confirm a healthy setup: one shows the live state of the daemon and your mesh, the other runs a real inference request.

Check the live state

waired status --observability

A healthy inference server looks like this:

Observability:
  Device:   dev-ubuntu  uptime: 1m23s
  Engine:   ready (model=qwen2.5-coder-7b-instruct, 0/4 slots used, inflight=0)
  Share:    enabled    Paused: no
  Mesh:     2 enrolled / 1 reachable / 1 ready
  Last:     -  (no inference requests yet)

What to look for:

• Engine: ready — the model is loaded and can serve requests. If it says not ready, the model may still be pulling (wait a few minutes) or the runtime may be down (waired runtimes status).
• Share / Paused — Share: enabled means peers can reach this engine; Paused: no means routing is active.
• Mesh — enrolled / reachable / ready peer counts. On a client, this is how you confirm your server shows up as reachable.

Add -o json for machine-readable output.

Run your first request

waired infer "say hi"

This calls your local engine directly, so it works regardless of mesh reachability — it’s the earliest way to confirm inference works. A greeting should come back within a few hundred milliseconds to a few seconds.

To see the routing decision (local vs. peer, model resolution, estimated latency) alongside the answer:

waired infer --explain "say hi"

Tip

If a request returns a 503, it usually means the engine is paused (waired resume) or not shared (waired inference share on). The Troubleshooting page maps each symptom to the command that fixes it.

What’s next

• Route your coding agent through your model: Coding agents.
• Connect a chat or OpenAI-compatible client: Chat clients.
• Pick a bigger or smaller model: Switch the bundled model.