gemini-models_skill

Overview

This skill provides direct access to Gemini 2.5 Pro and Flash family models through the SkillBoss API, with up to a 1M token context and multimodal input support. It removes the need for Google Cloud or Vertex AI configuration so you can quickly run long-context, multimodal workloads. The interface is optimized for processing long documents, videos, and large codebases with predictable pay-per-token pricing.

How this skill works

The skill sends chat completion requests to SkillBoss endpoints and selects the appropriate Gemini variant (2.5 Pro, 2.5 Flash, or Flash Lite) based on context length and latency/cost tradeoffs. It accepts text and multimodal inputs and returns model-generated summaries, analyses, or code-level insights. Authentication uses a single API key and no Google Cloud setup is required.

When to use it

• Summarizing or analyzing very long documents, books, or research corpora.
• Ingesting and extracting insights from large code repositories or multi-file projects.
• Processing multimodal assets like images, audio, or video for understanding or summarization.
• When you need a large context window (1M tokens) for complex reasoning across many documents.
• When you want lower, predictable token-based pricing without Vertex/Google Cloud setup.

Best practices

• Choose gemini-2.5-pro for the highest-context reasoning and complex document workflows.
• Use gemini-2.5-flash when you need faster responses at lower cost; use flash-lite for smallest budgets.
• Chunk extremely large inputs into logical sections and keep a pointer index to preserve coherence across calls.
• Provide clear system and user prompts to focus the model on the exact output format you need.
• Monitor token use and trim or compress inputs to control costs on high-volume workloads.

Example use cases

• Summarize a 500-page technical manual into chapter-by-chapter executive summaries.
• Analyze a large codebase to produce architecture diagrams, dependency maps, and refactor suggestions.
• Generate searchable indices and question-answering layers over a research paper collection.
• Transcribe and summarize long-form video interviews and extract action items.
• Run multimodal QA where images and text are combined for richer document understanding.