kunchenguid/axi

What it solves

AXI (Agent eXperience Interface) addresses the high overhead associated with AI agents interacting with external services. Traditional CLIs (built for humans) and structured tool protocols like MCP (Model Context Protocol) often consume excessive tokens and require more turns to complete tasks, increasing costs and latency.

How it works

AXI defines a new paradigm for building "agent-native" CLI tools. Instead of wrapping existing tools in a complex protocol, it establishes 10 design principles to optimize the interface between the agent and the tool. These principles focus on minimizing token consumption (using the TOON format for ~40% savings over JSON), truncating large content, providing pre-computed aggregates to reduce round trips, and ensuring structured errors and definitive empty states.

Who it’s for

Developers building AI agents and those creating tools for agents to use. It provides a framework for tool developers to make their services more "agent-ergonomic," reducing the cost and token budget of agentic workflows.

Highlights

• Token-Efficient Output: Uses the TOON format to significantly reduce token usage compared to JSON.
• Agent-Native Design: 10 specific design principles (e.g., minimal default schemas, content truncation, and contextual disclosure) to optimize for LLM interaction.
• Proven Performance: Benchmarks show higher success rates, lower average costs, and fewer turns compared to standard CLIs and MCP implementations for GitHub and browser automation.
• Extensive Ecosystem: A growing catalog of official and community-developed AXIs for services like GitHub, Chrome DevTools, AWS, Kubernetes, Slack, and Notion.