Routing Explainer

Run the terminal-native route explanation:

inferctl route code --prompt-file sample-task.txt --explain

From this directory:

./run.sh

Primary route selected:

selected: qwen3:8b on ollama
reason: primary model is available

candidates:
  role      backend          model                  status
  primary  ollama           qwen3:8b               selected_ready

prompt:
  15 estimated tokens / 8192 max context tokens

next: inferctl model qwen3:8b --json

Fallback selected:

selected: qwen3:8b on ollama_small (fallback)
reason: selected fallback because primary 'qwen-coder-32b.gguf' is unavailable

candidates:
  role      backend          model                  status
  primary  llamacpp_large   qwen-coder-32b.gguf    backend_unreachable
  fallback ollama_small     qwen3:8b               selected_ready

prompt:
  15 estimated tokens / 8192 max context tokens

warnings:
- W_BACKEND_UNREACHABLE: backend 'llamacpp_large' is unreachable
- W_FALLBACK_USED: routed to fallback 'qwen3:8b' because primary 'qwen-coder-32b.gguf' is unavailable

next: inferctl backends --filter llamacpp_large --json

route --explain is the human explanation of which configured local route inferctl would choose and why. route --json is the structured form for tools and tests. Use preflight when automation needs pass/fail readiness behavior before attempting work.

route --explain does not run inference. It answers which configured local route inferctl would choose for a task, why that route was selected, which candidates were considered, and which constraints shaped the decision. It does not judge answer quality, rewrite prompts, call the selected model, start backends, repair services, or warm models.

Use route --json when tools need the route decision object, candidate list, warnings, and commands in a machine-readable form. Use preflight when an automated workflow needs pass/fail readiness behavior before starting work. Use diff when the question is what changed between two captured control-plane states.