Show HN: ChatGemini – Easy tool, >50% cost saved with cache, multi(NL video edit (github.com)

ChatGemini_SakiTool (v1.0.1) is an open‑source, MIT‑licensed CLI toolkit that bundles Google Gemini text, image and video capabilities (including Veo 3.1 generation and a Flow Engine for NL-driven video editing) into a one‑line install / one‑command workflow. It supports Gemini 2.5 Pro/Flash/Flash‑8B/2.0, streaming output, long inputs (up to ~2M tokens), OCR, object detection, scene detection, subtitle generation, audio processing, and end‑to‑end media pipelines (mp4/mov/avi/webm). The release emphasizes usability (prompt‑toolkit TUI, history, tab completion), robust error handling/retry, checkpointing, scheduling/parallelism, and detailed performance/cost reporting. It requires a Gemini API key / Google AI billing and ships with a lightweight orthogonal vector DB (SQLite + NumPy) and CodeGemini embeddings for searchable code/context. The headline technical hook is its smart cache system: LRU + automatic cache creation at configurable thresholds (default 5k tokens) that the author claims saves 50–95% in token costs (examples show ~90% savings for repeat queries). That, plus fine‑grained "think" modes to allocate token budgets, makes the toolkit attractive for researchers and engineers prototyping multimedia LLM workflows where API spend and latency matter. Practical implications include much cheaper iteration on large-context, multi‑modal tasks and an easy local CLI for reproducible experiments — while still requiring responsible API/key management and awareness of cached sensitive content.