name	nexus-router
description	OpenClaw skill for Nexus Router — routes each chat turn to the best model using task classification, model capability scoring, and provider health. Use when you want automatic model selection instead of a fixed default.

Nexus Router Skill

Smart LLM router that picks the best model for each request based on:

Task type (coding, review, reasoning, summarization, vision, etc.)
Model capability scores
Provider health and quota status
Learned historical performance

Prerequisites

The Nexus Router Python server must be running before this skill works:

cd ${NEXUS_ROUTER_SRC:-/home/ubuntu/git/ghabs/nexus-router}
venv/bin/python -m src.server --port 7771

To run it as a background service, see systemd/nexus-router.service.

How it works

message → pre-signal extraction → heuristic classification
       → (optional LLM classifier for ambiguous cases)
       → scorer (task_fit × health × preference × learned × cost × speed)
       → selected_model + fallbacks + reason

Configuration (openclaw.json)

{
  "plugins": {
    "allow": ["nexus-router"],
    "entries": {
      "nexus-router": {
        "enabled": true,
        "config": {
          "routerUrl": "http://127.0.0.1:7771",
          "costProfile": "balanced",
          "useLlmClassifier": false,
          "debugMode": false,
          "minConfidence": 0.60
        }
      }
    }
  }
}

Config options

Key	Default	Description
`routerUrl`	`http://127.0.0.1:7771`	Nexus Router server URL
`costProfile`	`balanced`	cheap / balanced / premium
`useLlmClassifier`	`false`	Use LLM for ambiguous messages (costs a micro-call)
`debugMode`	`false`	Log routing decisions to console
`minConfidence`	`0.60`	Skip routing if classifier confidence is below this

CLI test

# Route a coding task
cd ${NEXUS_ROUTER_SRC:-/home/ubuntu/git/ghabs/nexus-router}
venv/bin/python -m src.cli route --task coding --complexity high --has-code

# Route a vision task
venv/bin/python -m src.cli route --task vision --has-image

# Check model stats after use
venv/bin/python -m src.cli stats

Supported task types

Type	Description
`coding`	Code generation, implementation, debugging
`code_review`	PR review, diff analysis
`reasoning`	Planning, strategy, architecture
`summarization`	Summaries, extraction, TL;DR
`fast_utility`	Quick answers, minor edits
`long_context`	Large documents, many files
`vision`	Images, screenshots, diagrams
`general_chat`	Conversational, general questions

Regenerate model registry

Run after openclaw update or when providers change:

cd ${NEXUS_ROUTER_SRC:-/home/ubuntu/git/ghabs/nexus-router}
openclaw models list --all --json > ~/.local/state/nexus-router/generated/openclaw-models.json
venv/bin/python src/generate_registry.py

name	nexus-router
description	OpenClaw skill for Nexus Router — routes each chat turn to the best model using task classification, model capability scoring, and provider health. Use when you want automatic model selection instead of a fixed default.