Token Saver Meta

"Token Saving for the Masses"

A meta-package that bundles token-saving tools across ALL layers of LLM token consumption — with zero configuration.

Why Token Saver Meta?

AI coding agents spend 40–60% of their token budget reading files and processing verbose output — not solving problems. Token Saver Meta installs a full stack of tools that slash consumption at every layer: agent behavior, code exploration, shell output, and tool schemas. No config. One install. All platforms.

Architecture

Hybrid across 4 layers:

graph TB
    subgraph LAYER4["Layer 4 — Power Tools • One-click enable"]
        L4["Schema compression · Cross-session memory<br/>Prompt optimization · Skill tuning<br/><b>~62% savings on tool schemas</b>"]
    end
    subgraph LAYER3["Layer 3 — Output Compression • Auto-install"]
        L3["Shell filtering · CLI optimization<br/><b>~70% savings on shell output</b>"]
    end
    subgraph LAYER2["Layer 2 — Code Intelligence • Auto-install"]
        L2["Code graph queries · Context mapping<br/>Repo packing<br/><b>~85% savings on file exploration</b>"]
    end
    subgraph LAYER1["Layer 1 — Agent Behavior • Instant, zero config"]
        L1["Prose terseness · Code minimalism<br/>Operational efficiency · Structured output<br/><b>~40% savings on agent output</b>"]
    end
    L1 --> L2 --> L3 --> L4

    style LAYER1 fill:#1a1a2e,stroke:#0f3460,color:#e0e0e0
    style LAYER2 fill:#16213e,stroke:#0f3460,color:#e0e0e0
    style LAYER3 fill:#0f3460,stroke:#533483,color:#e0e0e0
    style LAYER4 fill:#533483,stroke:#0f3460,color:#e0e0e0
    style L1 fill:#1a1a2e,stroke:#0f3460,color:#e0e0e0
    style L2 fill:#16213e,stroke:#0f3460,color:#e0e0e0
    style L3 fill:#0f3460,stroke:#533483,color:#e0e0e0
    style L4 fill:#533483,stroke:#0f3460,color:#e0e0e0

Covers all 7 token consumption types: exploration, shell output, agent output, prompt input, repeated knowledge, tool schemas, and instructions.

Quick Start

# npm
npx token-saver-meta

# Python
uvx token-saver-meta

# curl (single command, no package manager needed)
curl -fsSL https://raw.githubusercontent.com/Im-Busy/token-saver-meta/main/install.sh | sh

Full architecture: docs/architecture-v2.md

Multi-Platform Support

Supports 18 AI coding platforms out of the box:

Kilo · Claude Code · Codex · Cursor · Cline · Roo Code · Continue.dev · OpenCode · GitHub Copilot · Windsurf · Augment · FactoryAI · Gemini · Hermes · Kiro · Mastra · Pi

License

Apache-2.0