model-router
Intent-based model routing overlay for Hermes Agent. Use when a request should be mapped to a configured provider/model route such as image-prompt to Qwen 4B, debugging to Sonnet, cheap routine chat, or a user-defined route in skills.config.model-router.routes.
Resources
5Install
npx skillscat add zelray/hermes-model-router Install via the SkillsCat registry.
Model Router for Hermes Agent
This Hermes skill adds intent-based model-routing guidance on top of stock Hermes. Hermes already supports auxiliary.*, delegation.*, fallback_model, custom providers, /model, and per-run --provider --model overrides. This skill does not add a Hermes core feature; it reads user-editable route settings from skills.config.model-router and returns the safest command or session-switch recommendation.
Load the routing table
Read ~/.hermes/config.yaml and inspect:
skills:
config:
model-router:
default_route: routine_chat
execution_mode: advise_or_oneshot
routes: {}If the table is absent, ask the user to add the sample from assets/config.yaml.routed. Do not invent a root intent_routes: key.
Route selection policy
Prefer native auxiliary.* slots for built-in fixed tasks. Use this skill only for semantic user intent. Match the user request against each route's keywords and intent. If multiple routes match, prefer the more specific and higher-capability route. If no route matches, use default_route.
Execution policy
command_mode |
Behavior |
|---|---|
oneshot |
Build an isolated hermes -z "..." --provider <provider> --model <model> command. |
advise |
Recommend the exact /model provider:model or hermes chat --provider ... --model ... -q ... command. |
delegate |
Use Hermes delegation only when the task can be isolated as a subtask and the configured delegate model is appropriate. |
current_session |
Stay on the current/default session model. |
Always state selected route, provider, model, rationale, fallback route, and exact command. Never use /model ... --global unless the user explicitly asks to persist the change.
Optional helper
When terminal access is available, run:
python scripts/model_router.py --prompt "<user request>" --config ~/.hermes/config.yamlUse --execute only after the user confirms that an isolated hermes -z call is acceptable.