kunchenguid/axi

AXI – Agent eXperience Interface

What it is – AXI is a design framework and a set of concrete command‑line tools that are built for AI agents rather than for humans. It defines ten “agent‑ergonomic” principles (token‑efficient output, minimal schemas, explicit empty states, etc.) and provides a catalog of AXI‑wrapped CLIs (GitHub, Chrome DevTools, AWS, Docker, Notion, etc.) that follow those rules.

Why it matters – Large‑language‑model agents have a strict token budget when they call external services. Traditional CLIs and tool‑protocols (e.g., MCP) are verbose, require interactive prompts, and waste tokens on unnecessary JSON fields. AXI’s principles keep the data small, deterministic, and easy for an LLM to parse, which reduces cost and latency while improving reliability.

Key components

• 10 design principles – token‑efficient TOON format, tiny default schemas, explicit empty results, structured errors, ambient context, content‑first defaults, etc.
• AXI Catalog – a curated list of ready‑made AXI tools. Official examples include:
  • gh-axi – a token‑efficient wrapper around the GitHub CLI.
  • chrome-devtools-axi – browser automation via Chrome DevTools MCP.
  • lavish-axi, quota-axi, and many community‑contributed AXIs for Android, npm, SQLite, Slack, AWS, Docker, and more.
• Benchmarks – the README shows empirical results (Claude Sonnet 4.6) where AXI‑wrapped tools achieve higher success rates, lower cost, fewer turns, and faster runtimes compared with standard CLIs or MCP‑based approaches.

Typical workflow

1. Install an AXI tool (e.g., npm install -g gh-axi).
2. In an LLM‑driven script or prompt, invoke the tool just like any shell command.
3. The tool returns TOON (a compact, token‑saving format) with optional aggregates and next‑step suggestions.
4. The agent can parse the output, decide the next action, or present the data to a user.

Getting started

# Install the two flagship AXIs
npm install -g gh-axi
npm install -g chrome-devtools-axi

# Example usage from an LLM prompt
Use `gh-axi issue list --repo owner/repo --state open` to fetch open issues.
Use `chrome-devtools-axi navigate https://example.com` to drive a headless browser.

Add the tools to your agent’s knowledge base (e.g., in CLAUDE.md or AGENTS.md) so the model knows they exist.

Who should use it

• Developers building autonomous agents that need to interact with SaaS APIs, cloud services, or local tooling.
• Prompt engineers looking to reduce token consumption and improve reliability of tool calls.
• Anyone interested in a standardized, low‑overhead way for LLMs to perform real‑world actions.

Where to learn more

• Project website: https://axi.md
• Discord community for discussion and support.
• The principles.yaml and catalog.yaml files in the repo detail the full specification and the growing list of AXIs.

---

All information above is taken directly from the repository’s README; no additional features have been inferred.