Operations

This page covers day-to-day operations: checking fleet health, tailing logs, restarting components, and handling common issues.

Fleet health overview

From your client-front machine, get a summary of the full fleet:

pantheon status

Output:

Pantheon fleet status — 2026-05-30 10:14:22

  Iris          ● running   127.0.0.1:8080
  hermes        ● running   orchestrator.local:7677
  minotaur      ● running   orchestrator.local
  backend-01    ● running   qwen2.5:72b, llama3.1:8b
  backend-02    ○ offline   last seen 3h ago

Routing: 847 requests today — 612 local (72%), 235 Anthropic (28%)

Log tailing

Each component has a log subcommand. All accept --tail N for last N lines and stream live by default.

pantheon iris log               # routing decisions
pantheon minotaur log           # orchestrator dispatches
pantheon hermes log             # fleet coordination events
pantheon hydra log              # protocol translations (run on backend host)

Filter to errors only:

pantheon iris log --level error

Restarting components

pantheon iris restart
pantheon minotaur restart
pantheon hermes restart
pantheon hydra restart          # run on the backend host

Restart order if you need a full fleet restart:

1. Restart Ollama (if needed) on each backend host.
2. Restart hydra on each backend host.
3. Restart hermes on the orchestrator host.
4. Restart minotaur on the orchestrator host.
5. Restart Iris on the client-front machine.

Checking backend models

On the orchestrator host, check what models are loaded across all backends:

pantheon hermes backends

Output:

backend-01   qwen2.5:72b (loaded), llama3.1:8b (loaded)
backend-02   offline

On a backend host directly:

pantheon hydra models

Pulling new models

Pull a new model on a backend host, then notify hermes so the capability table updates:

ollama pull qwen2.5:14b
pantheon hydra reload           # re-registers loaded models with hermes

Common issues

Iris shows “fleet unreachable”

hermes is down or the orchestrator host is off the network. Check:

pantheon hermes status          # from the orchestrator host

If hermes is running but Iris can’t reach it, check firewall rules on port :7677.

Iris will fall back to Anthropic automatically if your routing policy has default = "auto" or default = "anthropic".

Requests fail with “no backend available”

No backend has the requested model class loaded. Check pantheon hermes backends and either:

• Pull the model on a backend host (ollama pull <model> + pantheon hydra reload), or
• Update your routing policy to allow Anthropic fallback for that model class.

hydra returns 400 for a model

The model name is not in hydra’s [model_map]. Add the mapping to /etc/pantheon/hydra.toml and run pantheon hydra restart.

Updating Pantheon

pantheon update                 # pulls latest components and restarts affected services
pantheon update --check         # check for updates without installing

Run pantheon update on each machine in the fleet. Components are backward-compatible within a minor version.