A weighted, policy-aware routing system for selecting the optimal AI model based on real-world performance, cost, and community sentiment.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install model-matrix
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 model-matrix 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 Model Matrix?
The Model Matrix skill is a sophisticated decision engine designed for developers using Openclaw Skills who need to manage multiple LLM providers efficiently. It employs a weighted scoring system to determine which model should handle specific task categories, such as complex coding or research, ensuring you always get the best results at the lowest possible price point.
By integrating this skill into your workflow, you can automate model selection based on a blend of real-world evaluation data, industry benchmarks, social sentiment, and operational costs. This ensures that your Openclaw Skills setup remains resilient and cost-effective even as the AI landscape rapidly evolves, allowing for seamless transitions between providers like Anthropic, OpenAI, and Google Gemini.
Model Matrix Use Cases
- Selecting the most cost-effective model for routine cron operations or basic coding tasks to save on API credits.
- Routing high-complexity enterprise discussions and architectural planning to premium models like GPT-5.3 Codex.
- Automating model fallbacks when specific providers are excluded or underperform based on recent benchmarks.
- Integrating real-time sentiment analysis from platforms like X and Reddit into your model selection logic for Openclaw Skills.
How Model Matrix Works
- The skill identifies the current task category, such as Research, Planning, or Creative Writing.
- It calculates a raw score for all available models using a weighted formula: 45% real task evals, 30% benchmarks, 20% sentiment, and 5% cost.
- Core policies are applied to check for provider exclusions or specific organizational constraints (e.g., auto-promoting alternatives if a primary provider is unavailable).
- The skill compares the score delta between the top candidates to ensure changes are only made when confidence is high and the improvement is material.
- It routes the request to the Effective #1 model and logs the decision in the daily scorecard.
Model Matrix Setup
To integrate the Model Matrix into your environment, you must add the skill to your active Openclaw Skills configuration. Use the CLI to initialize the skill:
openclaw install model-matrix
Once installed, verify your credentials for the routed providers (Gemini, GPT, Grok) are present in your environment variables to allow the matrix to switch routes dynamically.
Model Matrix Data Schema & Taxonomy
The Model Matrix organizes its performance tracking through a daily scorecard template. This allows Openclaw Skills to maintain a historical record of model efficiency across various dimensions:
| Attribute | Type | Description |
|---|---|---|
| Category | String | The task domain (e.g., Coding, Planning, Sentiment) |
| Raw Score | Integer | The calculated /100 score based on weighted inputs |
| Raw #1 | String | The highest-scoring model before policy constraints |
| Effective #1 | String | The final model selected after applying fallbacks |
| Confidence | Percentage | A metric indicating the reliability of the routing choice |
Model Matrix Advanced Features
- Multi-agent support: Automatically routes sub-tasks to specialized models (e.g., Gemini for images, GPT for logic) within a single Openclaw Skills workflow.
- Dynamic Sentiment Integration: Leverages the Grok ecosystem to adjust model scores based on real-time developer feedback from social platforms.
- High-Confidence Thresholds: Prevents model flapping by requiring a material score delta before switching routes between versions.
- Policy-Aware Auto-Promotion: Intelligent fallback logic that promotes the next best model if a specific provider is excluded by organizational policy.
SKILL.md
Loading
Related Openclaw Skills
Zettel Link
A semantic search and connection discovery tool for Obsidian vaults using local or cloud-based AI embeddings.
Weibo Timeline Monitor
Automates the monitoring and archiving of the Weibo Latest Posts timeline into organized, deduplicated Markdown files.
Claude Code Integration
A comprehensive integration for OpenClaw that brings Claude Code's advanced AI-assisted coding, documentation queries, and subagent management into your workflow.
Xiaohongshu Video Downloader
An automated utility for extracting and downloading high-quality videos from Xiaohongshu (Red) using Openclaw Skills.
Oblien Workspace Runtime
A secure, isolated Firecracker microVM environment providing a robust Internal API for AI agents to manage files, execute commands, and run interactive terminals.
Context Bridge
A sophisticated persistent memory layer that enables AI agents to resume complex projects with full context of past conversations and decisions.
Featured*
