An open-source LLM router that classifies prompts in milliseconds to slash API costs by routing to the most efficient model available.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install nadirclaw
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install nadirclaw using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is NadirClaw?
NadirClaw is a sophisticated LLM routing utility designed to drastically reduce expenditure by intelligently managing prompt distribution. By analyzing the complexity of a prompt in approximately 10ms, it determines whether a task requires a premium model like Claude 3.5 Sonnet or if it can be handled by a significantly cheaper or local model via Ollama. This capability makes it an essential tool for developers utilizing Openclaw Skills to scale their AI applications efficiently.
Functioning as an OpenAI-compatible proxy, NadirClaw integrates seamlessly into existing workflows. It provides a bridge between high-performance proprietary models and cost-effective local alternatives, ensuring that every token spent is optimized for both quality and price. For those managing high-volume AI agents, this skill provides the necessary infrastructure to maintain performance while protecting the bottom line.
NadirClaw Use Cases
- Slashing monthly AI API spend by 40-70% across providers like OpenAI and Anthropic.
- Automatically routing simple coding queries to local Llama models while reserving premium models for complex architecture.
- Ensuring high availability by implementing auto-retry fallbacks when primary providers hit rate limits.
- Maintaining conversational consistency through session persistence during multi-turn interactions.
How NadirClaw Works
- The user installs the package and configures API providers through an interactive wizard.
- The skill is onboarded to Openclaw Skills to act as a centralized provider proxy.
- The NadirClaw server is launched locally, exposing an OpenAI-compatible endpoint.
- Incoming prompts are analyzed by a lightweight classifier to determine required model capabilities.
- Prompts are routed to the optimal model based on the selected profile (eco, premium, or auto).
- Real-time cost savings and routing analytics are logged and viewable via a terminal dashboard.
NadirClaw Setup
Install the core package via pip:
pip install nadirclaw
Initialize the configuration wizard:
nadirclaw setup
Integrate directly with Openclaw Skills:
nadirclaw openclaw onboard
Start the routing service:
nadirclaw serve --verbose
NadirClaw Data Schema & Taxonomy
NadirClaw organizes its routing logic and analytics through a structured local taxonomy to ensure speed and data privacy.
| Component | Details |
|---|---|
| Routing Profiles | Predefined behaviors including auto, eco, premium, and reasoning. |
| Provider Config | YAML-based storage for API keys and endpoint URLs for OpenAI, Google, and Ollama. |
| Analytics Data | Tracks cost per request, token throughput, and routing efficiency. |
| Session Logs | Detailed history of routing decisions accessible via nadirclaw report. |
NadirClaw Advanced Features
- Agentic task detection that automatically forces premium model usage when complex tool-calling is identified.
- Context-window filtering that auto-swaps models when conversation history exceeds specific thresholds.
- Rate limit fallback mechanisms that trigger auto-retries on 429 errors from primary providers.
- Support for any LiteLLM provider, enabling a massive range of backend model options within Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Zero Trust Security Protocol
A security-first behavioral framework that enforces strict verification for all external AI agent operations and sensitive data handling.
PrusaLink CLI
A secure, curl-based command-line interface for managing Prusa 3D printers locally through PrusaLink.
Workspace Organization
Automated maintenance and structural standardization tool to prevent directory entropy and ensure healthy AI agent deployments.
Skill Vetter
A security-first vetting protocol designed to audit AI agent skills for malicious code, credential theft, and unauthorized data exfiltration before installation.
Humanizer Enhanced
A professional-grade content refinement tool that detects and eliminates 34 distinct AI writing patterns while injecting personality and soul into drafts.
Ralph Quick Security Auditor
A high-speed security auditor that performs a 10-step spot-check for vulnerabilities in under 10 minutes.
Featured*
