openrouter-toolkit

What This Skill Does

The openrouter-toolkit is a comprehensive management layer for your OpenClaw agent, designed to orchestrate LLM interactions through the OpenRouter API. At its core, the toolkit functions as an intelligent load balancer and financial controller. It automates model selection based on task-specific heuristics, ensuring that you aren't using an expensive flagship model for simple text classification, while ensuring that complex coding or reasoning tasks receive the high-parameter models they require. Beyond routing, it provides a persistent audit trail for API expenditures via a local SQLite database, allows for seamless fallback orchestration if a provider experiences downtime, and enables native model comparison benchmarks without leaving your terminal environment.

Installation

To integrate this skill into your workflow, ensure your environment meets the Python 3.10+ requirement. Execute the following installation command in your terminal:

clawhub install openclaw/skills/skills/aiwithabidi/openrouter-toolkit

Once installed, you must configure your credentials. Locate your OpenRouter API key and export it as an environment variable: export OPENROUTER_API_KEY='your_key_here'. The toolkit automatically initializes the local SQLite database at {baseDir}/data/openrouter_costs.db upon the first request.

Use Cases

• Automated Cost Optimization: Developers can set strict monthly budget thresholds to prevent accidental overspending on high-token-count models during recursive agent loops.
• Resilient Pipeline Construction: Build agents that automatically retry failed requests on cheaper, faster fallback models to ensure continuous uptime.
• A/B Testing Prompts: Rapidly compare the output quality of different providers (e.g., Anthropic vs. OpenAI) for the same system prompt to fine-tune agent personality.
• Developer Productivity: Use the command-line interface to quickly query the current best models for specific categories like 'code' or 'creative' without navigating browser interfaces.

Example Prompts

1. "OpenClaw, use the openrouter-toolkit to route my current coding task to the best available model for complex logic."
2. "Compare the output of Claude 3.5 Sonnet and GPT-4o-mini for my current creative writing prompt and show me the cost difference."
3. "Set a monthly budget alert of $20.00 in the openrouter-toolkit and check my current spend for the week."

Tips & Limitations

When using the toolkit, prioritize the 'fast' and 'cheap' routing tasks for background processes and non-critical operations to maximize your API budget. Be aware that the fallback chain adds slight latency as it waits for a primary model timeout before triggering subsequent attempts. The tool relies on local SQLite storage, so ensure your agent has write permissions in the {baseDir}/data/ directory. Note that the toolkit pulls live data from the OpenRouter API; therefore, internet connectivity is mandatory for accurate model list updates and budget verification checks.