A policy-aware Terminal User Interface (TUI) for working with Google’s Gemini platform, combining a chat-forward experience with safe tool orchestration.

ADH CLI

ADH CLI is a policy-aware Terminal User Interface (TUI) for working with Google’s Gemini / ADK platform. It combines a chat-forward experience with tool orchestration, safety checks, and human-in-the-loop confirmation so you can explore ideas safely from your terminal.

Highlights

• Policy-aware orchestration – Every tool call flows through a policy engine (PolicyDecision, safety pipeline, audit logging) before execution.
• Human-in-the-loop tooling – The tool execution manager shows pending/active runs, supports confirmation and cancellation, and keeps a local history.
• Optional Google web tools – Gemini’s built-in web search and URL context (requires Python 3.10+).
• Configurable agents – Agents are defined in Markdown (adh_cli/agents/) and loaded at runtime with variable substitution and model overrides.
• Themed Textual UI – Dark/light themes, command palette integrations, clipboard export, and keyboard-friendly chat navigation.
• XDG-compliant storage – Config, policies, audit logs, and backups live under ~/.config/adh-cli/ via ConfigPaths.
• Well-tested codebase – 300+ pytest cases cover core logic, UI widgets, policies, safety checks, and services.

Quick Start

Prerequisites

• Python 3.10 or newer
• uv for fast, reproducible environments (pip works too)

Clone & Install

git clone https://github.com/allenhutchison/adh-cli.git
cd adh-cli

# Create and activate a virtual environment (uv recommended)
uv venv
source .venv/bin/activate  # Windows: .venv\Scripts\activate

# Install the project with development extras
uv pip install -e '.[dev]'
# or use the task helper
task install-dev

First Run

# Provide a Claude API key via env or .env
export GOOGLE_API_KEY="your-api-key"  # or GEMINI_API_KEY

# Launch the TUI
adh-cli              # console script entry point
# or use the task wrapper
task run

# Enable Textual dev tools
task dev