Language: English | 简体中文

Airymax CLI

Official command-line interface for the Airymax AI Agent Runtime Platform. One of the leaf repositories aggregated by the sdk management repo. Built on top of the Airymax Rust SDK (agentrt-rs).

---

Overview

The Airymax CLI (agentrt) is a Rust-built command-line tool for operating the Airymax runtime from the terminal. It covers the full agent lifecycle: project scaffolding, component creation, agent execution, configuration management, LLM provider management, prompt template management, marketplace search & install, deployment operations, and database migrations.

Runtime commands talk to the Airymax Gateway over HTTP; the CLI internally drives the same double-layer SDK architecture (Cognition / Safety / Tool / Chat) used by the language SDKs, so the CLI is a first-class runtime tenant that exercises the same code paths an agent application would.

Double-Layer API Architecture (consumed)

The CLI is a consumer of the Airymax Rust SDK rather than a provider of new resource clients. Its run, llm, market, and deploy subcommands translate user intent into calls against the four nested resource clients exposed by AgentRTClient:

agentrt <command>
   └── AgentRTClient (from agentrt-rs)
       ├── CognitionClient   # used by `agentrt run`
       ├── SafetyClient      # used by policy / audit flows
       ├── ToolClient        # used by tool invocation & market install
       └── ChatClient        # used by interactive chat sessions

Directory Structure

cli/
├── src/
│   ├── main.rs              # Entry point + CLI routing (clap)
│   ├── client.rs            # Gateway HTTP client
│   ├── templates.rs         # Project / component template generation
│   └── commands/
│       ├── mod.rs           # Command module exports
│       ├── init.rs          # agentrt init — project scaffolding
│       ├── create.rs        # agentrt create — agent/tool/plugin/prompt/skill
│       ├── run.rs           # agentrt run — run an agent (interactive / one-shot)
│       ├── config_cmd.rs    # agentrt config — configuration management
│       ├── llm.rs           # agentrt llm — LLM provider management
│       ├── prompt.rs        # agentrt prompt — prompt template management
│       ├── market.rs        # agentrt market — marketplace search & install
│       ├── deploy.rs        # agentrt deploy — deployment & status
│       └── db.rs            # agentrt db — database migration management
├── Cargo.toml               # Crate manifest (agentrt-cli, binary: agentrt)
└── README.md                # This file

Upstream & Downstream Dependencies

Upstream

• Airymax Rust SDK (agentrt-rs): Provides the typed AgentRTClient and the four nested resource clients used to talk to the runtime.
• Runtime: Connects to a running Airymax / AgentRT instance (gateway_d / Gateway HTTP API) over HTTP and JSON-RPC 2.0.
• Configuration: Resolved from CLI flags, then environment variables (AGENTRT_ENDPOINT, AGENTRT_API_KEY), then a http://127.0.0.1:18789 default.

Downstream

• Developers / operators: The primary human-facing surface for driving the runtime from a shell, CI job, or ops runbook.
• Shell completion: Generates completion scripts for bash/zsh/fish/etc.

Command Reference

Command	Description	Needs Gateway
agentrt init <name>	Initialize a new Airymax project	✗
agentrt create agent|tool|plugin|prompt|skill <name>	Scaffold a component	✗
agentrt run [prompt]	Run an agent (interactive / one-shot)	✓
agentrt config show|set|validate|reload	Configuration management	△
agentrt llm list|test|cost	LLM provider management	✓
agentrt prompt list|show|tune|ab-test	Prompt template management	✗
agentrt market search|install|publish	Marketplace search & install	✓
agentrt deploy deploy|status|logs	Deployment & operations	✓
agentrt db status|migrate|rollback|new	Database migration management	✗
agentrt completion <shell>	Generate shell completion	✗

Installation

From source

cd cli
cargo build --release
# Binary: ./target/release/agentrt

Shell completion

source <(agentrt completion bash)

Requirements: Rust edition 2021 (stable toolchain). Runtime dependencies: clap 4.5 (CLI framework), reqwest 0.12 (HTTP), tokio 1 (async runtime), serde / serde_json / serde_yaml (serialization), thiserror / anyhow (errors), colored / indicatif (terminal output), chrono, dirs, url, urlencoding, clap_complete.

Usage

Project scaffolding

agentrt init my-agent-project
cd my-agent-project

agentrt create agent my-agent
agentrt create tool web-scraper
agentrt create plugin logging-plugin

Run an agent

# Interactive mode
agentrt run

# One-shot
agentrt run "Analyze this sales data"

# With explicit agent config and model
agentrt run --agent-file agents/custom.agent.yaml --model gpt-4

Configuration management

agentrt config show
agentrt config set llm.providers.openai.api_key sk-xxxx
agentrt config validate
agentrt config reload

Prompt templates

agentrt prompt list
agentrt prompt show intent_classify
agentrt prompt tune intent_classify --dataset ./data.jsonl
agentrt prompt ab-test intent_classify --baseline v1 --candidate v2

Deployment & operations

agentrt deploy deploy --target docker
agentrt deploy status
agentrt deploy logs --lines 100

Build & Test

cargo build --release
cargo test
./target/release/agentrt --help

Branch Strategy

This leaf repository is developed on feature/official-hubs-01. The aggregating sdk management repo stays on main.

License

Dual-licensed under AGPL v3 + Apache 2.0 (SPDX: AGPL-3.0-or-later OR Apache-2.0). See LICENSE for the full text.

Copyright (c) 2025-2026 SPHARX Ltd. All Rights Reserved.