Route image understanding requests to the cheapest available backend.
npx skills add https://github.com/Lingtai-AI/lingtai --skill vision-lingtai-ai
Please help me install this Agent Skill. Skill: vision Source: https://github.com/Lingtai-AI/lingtai/tree/main/tui/internal/preset/skills/vision Command: npx skills add https://github.com/Lingtai-AI/lingtai --skill vision-lingtai-ai
This skill routes image understanding requests to the most cost-efficient available backend, automatically selecting between a built-in vision tool, the minimax-cli path, or a local Hugging Face VLM.
Instruct the agent to route an image to the cheapest available understanding path.
High-intent search queries and answers about installing and using this skill.
To route image understanding tasks to the cheapest vision model, this skill applies a deterministic decision tree that selects between a built-in vision tool, MiniMax CLI, or a local Hugging Face VLM based on environment availability.
For offline image analysis fallbacks, the router automatically switches to a local Hugging Face VLM when the primary built-in vision tool and MiniMax CLI paths are unavailable, ensuring reliable results.
Yes, you can use a local Hugging Face VLM alongside MiniMax for image OCR and description. The skill dynamically selects the MiniMax CLI path if a coding-plan key exists, or falls back to the local VLM.
Yes, running the local VLM path requires specific dependencies including transformers, torch, pillow, einops, and qwen-vl-utils to process and understand the image data offline.
If your agent fails to select the correct vision path for image critique, it is likely because the deterministic decision tree could not detect the required environment signals or the specific coding-plan key for fallbacks.
You need dynamic path selection for processing image requests when an agent must describe, OCR, or critique images with minimal latency and cost, adapting automatically to available backend tools.
founderphantomRun multimodal image understanding and conversation with a vision-language model.
LuminaVaultExtract text from WhatsApp images with Tesseract OCR and save to Obsidian.
vanman2024Validate OpenRouter API keys, models, routing, and environment configurations.
vinitgirdharRoute LLM tasks to models by complexity with cost tracking and retries.
