kunchenguid/axi

What it solves

AI agents often struggle with existing interfaces like standard CLIs (built for humans) or structured protocols like MCP, which can be token-heavy and inefficient. AXI provides a set of design principles for creating "agent-native" CLI tools that minimize token usage and reduce the overhead agents face when interacting with external services.

How it works

AXI is a framework of 10 design principles that prioritize token budget as a primary constraint. It encourages the use of the TOON format for token savings, minimal default schemas, content truncation with escape hatches, and pre-computed aggregates to eliminate unnecessary round trips. Developers can implement these principles to create AXI-compliant tools, and the project provides a "skill" (a detailed guide) that coding agents can use to scaffold and build new AXIs.

Who it’s for

Developers building AI agents and tool-makers who want to create interfaces that are more ergonomic, cost-effective, and reliable for LLMs to use.

Highlights

• Token-Efficient Output: Uses TOON format to achieve approximately 40% token savings over JSON.
• Agent-Native Design: Focuses on removing ambiguity (e.g., definitive empty states) and reducing turns (e.g., contextual disclosure).
• Proven Performance: Benchmarks show higher success rates, lower costs, and fewer turns compared to standard CLIs and MCP implementations for GitHub and browser automation.
• Extensive Catalog: A growing ecosystem of official and community-built AXIs for services like AWS, Docker, Kubernetes, Slack, and Notion.