nanoclaw_flexagents: Skill for Claude Code

Add Model Endpoint

Connect a new model provider so its models appear in the /model command.

Phase 1: Pre-flight

Ensure custom models are enabled

ls src/runtime/codex-runtime.ts 2>/dev/null && echo "READY" || echo "NOT_READY"

If NOT_READY, tell the user to run /add-custom-models first.

Phase 2: Select endpoint type

AskUserQuestion: What type of model endpoint?

1. OMLX (local, Apple Silicon) — Optimized for Mac. Persistent KV cache, tool calling, MCP support. Install: brew install omlx
2. Ollama (local) — Easy to use. Install: brew install ollama or https://ollama.com
3. Remote URL — Any OpenAI-compatible endpoint (Together AI, Groq, HuggingFace, Fireworks, self-hosted vLLM, etc.)

Phase 3: Configure based on selection

OMLX

1. Check if OMLX is installed and running:

curl -sf http://localhost:8000/v1/models 2>/dev/null | head -5

If not running, guide installation:

brew tap jundot/omlx https://github.com/jundot/omlx
brew install omlx
brew services start omlx

1. Check what models are available:

curl -sf http://localhost:8000/v1/models | python3 -c "import sys,json; [print(m['id']) for m in json.load(sys.stdin).get('data',[])]"

If no models, suggest downloading one:

# Models are downloaded on first use, or pre-download via OMLX admin UI at http://localhost:8000/admin

1. Add to .env:

grep -q 'OMLX_URL' .env || echo 'OMLX_URL=http://localhost:8000/v1' >> .env

1. Add models to src/config.ts under AVAILABLE_MODELS.local. Use the model IDs from the curl output above. Example:

local: [
  { id: 'mlx-community/Llama-3.1-8B-Instruct', name: 'Llama 3.1 8B (local)' },
],

Ollama

1. Check if Ollama is running:

curl -sf http://localhost:11434/api/tags 2>/dev/null | head -5

If not running: ollama serve or check https://ollama.com for installation.

1. List available models:

ollama list

If none installed, suggest:

ollama pull llama3.1        # General purpose
ollama pull codellama       # Code focused