hermesmade

HermesMade is a collection of standalone command-line interface (CLI) tools designed to address specific frustrations reported by AI users, particularly those found on Reddit. The project emphasizes a data-driven approach, identifying common pain points such as aggressive AI censorship, silent model degradation, opaque API pricing, and complex local LLM setup. The repository includes six primary tools: 1. **prompt-inspector**: Analyzes prompts for censorship risks by detecting trigger words across categories like violence, politics, and adult content. It suggests rewrite strategies (e.g., academic prefixes) and provides uncensored presets for local LLMs like Gemma, DeepSeek, and Llama. 2. **model-watch**: Monitors AI model quality by running standardized tests for reasoning, coding, and hallucination. It tracks score history over time and alerts users to significant degradation compared to historical baselines. 3. **api-cost**: Compares pricing across 18 models from providers like OpenAI, Anthropic, and DeepSeek. It recommends cost-effective models for specific scenarios (coding, chat, writing) and includes a spending tracker. 4. **llm-deploy**: Assists with local LLM setup by detecting hardware specs (RAM, GPU) and recommending suitable models and installation commands (e.g., for Ollama). 5. **code-inspector**: Scans AI-generated code for quality issues, such as hardcoded API keys or mutable defaults, providing a review score. 6. **task-cost-estimator**: Estimates costs for specific AI tasks. Installation is supported via a one-command script or individual pip installations for each tool. The project is licensed under MIT and structured as a monorepo with separate directories for each tool.