Skip to main content
WES-GBeta
  • Guidelines
  • About
  • Contribute
  • Get The Badge

Filter by topic

Governance

  • 0.1 Documented Policy
  • 0.2 Essential Tracking and Ads

AI

  • 8.1 Model Selection
  • 8.2 Context Management & Token Usage
  • 8.3 FM (Foundational Model)-as-a-Service & Shared Infrastructure
  • 8.4 Prompt Efficiency
  • 8.5 Cache and Reuse
  • 8.6 Loop & Iteration Management

UX / Design

  • 1.2 Rich Content
  • 1.3 Use of Images
  • 1.4 Awareness
  • 1.5 Data Transfer and Emissions
  • 1.6 Carbon Aware Design
  • 1.7 Colour Scheme
  • 1.8 User Retention

Images

  • 2.1 File Formats
  • 2.2 Image Optimisation
  • 2.3 Image Resolution
  • 2.4 Browser Cropping

Video

  • 3.1 Autoplay
  • 3.2 Script Loading
  • 3.3 Streaming Resolution

Content

  • 4.1 Content Audit
  • 4.2 Digital First
  • 4.3 Easy Access
  • 4.4 Descriptive Headings

Fonts

  • 5.1 Font Variations
  • 5.2 File Formats
  • 5.3 Clean Files
  • 5.4 System Fonts

Web Development

  • 6.1 Lazy Loading Images
  • 6.2 Responsive Design
  • 6.3 Modular Design
  • 6.4 Minification
  • 6.5 Templates
  • 6.6 Asset Loading
  • 6.7 Analytics
  • 6.8 Data Minimisation
  • 6.9 Stylesheet
  • 6.10 Text Compression
  • 6.11 Carbon Aware Development
  • 6.12 API Efficiency

Development Operations

  • 7.1 Bad Robots
  • 7.2 Dev Environments
  • 7.3 Dataset
  • 7.4 Site Architecture
  • 7.5 Caching
  • 7.6 Pipeline Code
  • 7.7 Dependency Patching
  • 7.8 Green Hosting

8.2 Context Management & Token Usage

Minimise what you send to and generate from an AI system, every token and every tool or MCP call, carries a real compute cost.

Easy to implement

High impact score

Introduction:

Input tokens, output tokens and tool-call payloads all have real energy behind them. This is worth being deliberate about with MCP (Model Context Protocol) tool calls in particular: each one typically returns a full schema definition plus a full response payload back into context, and chaining several calls across an agentic task compounds that quickly. Favour targeted retrieval over stuffing entire documents into context, trim conversation history once it’s no longer relevant to the task at hand, and treat every additional tool call as a cost to be justified rather than a free convenience.

Resources:

EcoLogits tracks the energy consumption and emissions of GenAI API calls (including Anthropic and OpenAI) directly in your code.

← Previous Next →

© 2026

Sustainable Website Design by: Kyan, development: Complex Creative & Studio 24

Privacy Policy Cookie Policy