VL-JEPA inspired pipeline — compress images/text locally via Ollama, send compact payloads to any LLM API. Cut token costs by ~80%.
Skill Claude CodeCodex
Compress images, long prompts, conversations, and RAG documents locally via Ollama before consuming LLM tokens. Reduces token usage by 80% on images and 85% on text. Use proactively when: the user uploads images, processes long context, works with multiple RAG sources, or when approaching token limits.
Skill Claude CodeCodex
Compress images, long prompts, conversations, and RAG documents locally via Ollama before consuming LLM tokens. Reduces token usage by 80% on images and 85% on text. Use proactively when: the user uploads images, processes long context, works with multiple RAG sources, or when approaching token limits.