Snippy

Snippy

Snippy orchestrates Anthropic's Claude to cut cost by intelligently routing tasks to the most efficient sub-models.

Features

Intelligent Routing: Uses a local Ollama model to determine the optimal Claude sub-model for each task
Cost Optimization:
Downshift savings: Using cheaper models for subtasks
Cache reuse: Leveraging prompt caching to reduce costs
Free routing: Avoiding expensive API calls with local routing
Parallel Processing: Subtasks run in parallel for faster execution
Prompt Cache Preservation: Maintains Claude's prompt cache efficiency
Visual Feedback: Savings tracking in the status bar and collapsible panel
**

How It Works

**
User selects a main Claude model per session
For each task, a local Ollama router proposes a plan:
Which Claude sub-models to use
Whether to split into parallel subtasks
User can edit the proposed plan
Approved subtasks run as stateless, one-shot Claude calls
Results are appended to the main conversation to maintain cache efficiency
Savings are tracked and displayed

Setup

Install the Snippy extension
Configure your Anthropic API key via "Snippy: Set API Key"
Set up Ollama locally with a router model (default: llama3.2:3b)
Select your main Claude model in settings
Savings Tracking
Snippy tracks three buckets:

Downshift Savings: Cost reduction from using cheaper sub-models
Prompt Cache Reuse: Savings from reusing cached prefixes
Free Local Routing: Cost avoided by using local Ollama instead of API calls
**

Configuration

**
snippy.mainModel: Main Claude model for the session
snippy.maxTokens: Maximum tokens for Claude responses
snippy.enablePromptCache: Enable prompt cache optimization
snippy.ollama.baseUrl: Base URL for Ollama server
snippy.ollama.routerModel: Ollama model for routing