What problem does it solve? Managing a limited Agnes AI quota (1,500 requests per 5-hour sliding window) requires strict routing so that only image and video generation consume Agnes tokens, while text, reasoning, code, and vision tasks use DeepSeek instead. ## Core Features & Use Cases - 2-Tier Request Routing: Forces image generation (text-to-image, image-to-image) and video generation (text-to-video, image-to-video) through Agnes, and sends everything else to DeepSeek. - Quota Monitoring: Integrates with a cronjob-based API status monitor that alerts when quota drops below 300 (warning) or 100 (critical). - Modular Key Management: Stores provider keys separately in ~/.hermes/env/ files (agnes.env, deepseek.env) sourced automatically via .bashrc. - Use Case: When generating a 5-second animation, the router sends the text-to-image step to Agnes Image 2.1 Flash, feeds the resulting public URL into Agnes Video V2.0 for image-to-video, and polls the task status until completion. ## Quick Start Route my request to generate an image with Agnes and handle all my text and code questions with DeepSeek.