Crate aether_cli 

aether-agent-cli

Binary package containing Aether’s two runnable entrypoints:

• aether-acp — Agent Client Protocol (ACP) server for editor/IDE integration (e.g. Zed)
• aether — Headless CLI for single-prompt usage

Table of Contents

• Quick Start
  • Build
  • Run the CLI
  • Run the ACP server
• Choosing a Model
• Editor Integration (ACP)
  • Zed
• MCP Configuration
• Slash Commands
• Settings
  • Agents (Modes and Sub-agents)
• Logs

Quick Start

Build

From the workspace root:

cargo build --release -p aether-agent-cli

Binaries will be at target/release/aether-acp and target/release/aether.

Run the CLI

cargo run -p aether-agent-cli --bin aether -- -m anthropic:claude-sonnet-4-5-20250929 "Refactor auth module"

Run the ACP server

cargo run -p aether-agent-cli --bin aether-acp -- --model anthropic:claude-sonnet-4-5-20250929 --mcp-config mcp.json

Choosing a Model

Aether supports multiple LLM providers using a provider:model string format:

Provider	Example	Env var required
Anthropic	anthropic:claude-sonnet-4-5-20250929	ANTHROPIC_API_KEY
OpenRouter	openrouter:moonshotai/kimi-k2-thinking	OPENROUTER_API_KEY
ZAI	zai:GLM-4.6	ZAI_API_KEY
Ollama	ollama:llama3.2	None (local)
Llama.cpp	llamacpp	None (local)

Editor Integration (ACP)

Zed

Add to your Zed settings.json (Main Menu -> “Open Settings File”):

{
  "agent_servers": {
    "Aether Agent": {
      "command": "/path/to/aether/target/release/aether-acp",
      "args": [
        "--model",
        "zai:GLM-4.6",
        "--mcp-config",
        "/path/to/aether/mcp.json"
      ],
      "env": {
        "RUST_LOG": "debug",
        "ZAI_API_KEY": "your-api-key-here"
      }
    }
  }
}

Then open the Agent Panel and select “New Aether Agent Thread”.

Important: Update the paths and configuration:

• command: Full path to your built aether-acp binary
• --mcp-config: Path to your MCP configuration file
• Set the appropriate API key env var for your model provider

MCP Configuration

The mcp.json file configures MCP tool servers:

{
  "servers": {
    "coding": {
      "type": "in-memory",
      "args": ["--rules-dir", ".aether/skills", "--rules-dir", ".claude/rules"]
    },
    "skills": {
      "type": "in-memory",
      "args": [
        "--dir", ".aether/skills",
        "--dir", ".claude/skills",
        "--notes-dir", ".aether/notes"
      ]
    }
  }
}

• coding — Filesystem tools (read, write, bash, etc.) + optional automatic read rules from configured --rules-dir paths
• skills — Slash commands and reusable prompts from configured prompt directories

Slash Commands

Create markdown files in ~/.aether/commands/ to define custom slash commands.

Example ~/.aether/commands/plan.md:

---
description: Create a detailed implementation spec for a task
---

You are an expert software architect. Create a comprehensive technical specification.

# Task
$ARGUMENTS

Parameter syntax:

• $ARGUMENTS — Full argument string (e.g., /plan add user auth -> “add user auth”)
• $1, $2, $3 — Positional arguments

Settings

Project-level agent configuration is centralized in .aether/settings.json in your project root. This file defines agents (modes and sub-agents), prompts, and MCP server configuration.

Agents (Modes and Sub-agents)

Define agents with specific model, prompts, and tool configurations:

{
  "agent": "planner",
  "prompts": [".aether/prompts/shared.md", "AGENTS.md"],
  "mcps": [".aether/mcp.json"],
  "agents": [
    {
      "name": "planner",
      "description": "Planner optimized for decomposition and sequencing",
      "model": "anthropic:claude-sonnet-4-5",
      "reasoningEffort": "high",
      "userInvocable": true,
      "agentInvocable": true
    },
    {
      "name": "researcher",
      "description": "Read-only research agent",
      "model": "anthropic:claude-sonnet-4-5",
      "userInvocable": false,
      "agentInvocable": true,
      "prompts": [".aether/prompts/researcher.md"],
      "mcps": [".aether/researcher-mcp.json"],
      "tools": {
        "allow": ["coding__grep", "coding__read_file", "coding__glob"],
        "deny": []
      }
    },
    {
      "name": "coder",
      "description": "Fast coding agent",
      "model": "deepseek:deepseek-chat",
      "userInvocable": true,
      "agentInvocable": false,
      "prompts": [".aether/prompts/coder.md"]
    }
  ]
}

• agent — Optional default user-invocable agent name.
• Top-level prompts — Ordered default prompt sources used by agents that do not define their own prompts. File paths can be written as strings; typed objects support { "type": "text", "text": "..." }, { "type": "file", "path": "..." }, and { "type": "glob", "pattern": "..." }.
• Top-level mcps — Ordered default MCP config sources used by agents that do not define their own mcps. File paths can be written as strings; typed objects support { "type": "file", "path": "...", "proxy": false } and inline { "type": "inline", "servers": { ... } } entries.
• Agent prompts — Optional ordered prompt sources that override top-level prompts for that agent. Supports the same string shorthand and typed objects as top-level prompts.
• Agent mcps — Optional ordered MCP config sources that override top-level mcps for that agent. Supports the same string shorthand and typed objects as top-level mcps.
• userInvocable: true — Agent appears as a mode option in ACP clients (e.g., Wisp’s Shift+Tab)
• agentInvocable: true — Agent can be spawned as a sub-agent
• tools — Filter which MCP tools the agent can use (optional). Supports allow (allowlist) and deny (blocklist) with trailing * wildcards. If both are set, allow is applied first, then deny removes from the result. Omit or leave empty to allow all tools.

Logs

Logs are written to --log-dir (default: /tmp/aether-acp-logs/). Control verbosity with the RUST_LOG environment variable.

Re-exports

pub use acp::map_mcp_prompt_to_available_command;

Modules

acp

agent

error

headless

mcp_config_args

prompt

resolve

runtime

sandbox

settings_args

show_prompt