aria-llm

Access multiple LLM providers via LiteLLM routing for text generation and chat.

Model Priority (Feb 2026)

•Local MLX (qwen3-mlx) - FREE, fastest, no rate limits
•OpenRouter FREE - No cost, may have rate limits
•Kimi (paid) - Last resort, costs money!

Available Models

Model	Provider	Context	Best For
`qwen3-mlx`	Local MLX	32K	Primary - Fast local
`trinity-free`	OpenRouter	128K	Agentic, creative
`qwen3-coder-free`	OpenRouter	262K	Code generation
`chimera-free`	OpenRouter	164K	Reasoning (fast)
`qwen3-next-free`	OpenRouter	262K	RAG, tools
`glm-free`	OpenRouter	131K	Agent-focused
`deepseek-free`	OpenRouter	164K	Deep reasoning
`nemotron-free`	OpenRouter	256K	Long context
`gpt-oss-free`	OpenRouter	131K	Function calling
`kimi`	Moonshot	256K	PAID - Avoid!

Usage

bash

exec python3 /root/.openclaw/workspace/skills/run_skill.py llm <function> '<json_args>'

Functions

generate

Generate text from a prompt using specified model.

bash

exec python3 /root/.openclaw/workspace/skills/run_skill.py llm generate '{"prompt": "Explain quantum computing simply", "model": "qwen3-mlx", "temperature": 0.7}'

chat

Multi-turn conversation with message history.

bash

exec python3 /root/.openclaw/workspace/skills/run_skill.py llm chat '{"messages": [{"role": "user", "content": "Hello!"}], "model": "qwen3-mlx"}'

analyze

Analyze text for sentiment, topics, or custom analysis.

bash

exec python3 /root/.openclaw/workspace/skills/run_skill.py llm analyze '{"text": "I had a great day today!", "analysis_type": "sentiment"}'

Model Selection Guide

code

IF task = code_generation OR code_review:
    USE qwen3-coder-free (262K context, optimized for code)
ELIF task = complex_reasoning:
    USE chimera-free (fast reasoning) OR deepseek-free (deep reasoning)
ELIF task = creative_writing OR roleplay:
    USE trinity-free (best for creative)
ELIF task = long_context OR RAG:
    USE qwen3-next-free (262K) OR nemotron-free (256K)
ELSE:
    USE qwen3-mlx (default local, fastest)

API Configuration

Required environment variables:

•OPEN_ROUTER_KEY - OpenRouter API key (for FREE models)
•MOONSHOT_KIMI_KEY - Moonshot API key (paid fallback)
•OLLAMA_URL - Ollama endpoint (backup local)

Python Module

This skill wraps /root/.openclaw/workspace/skills/aria_skills/llm.py