localagentcli 0.1.0

A production-grade, local-first AI CLI

LocalAgentCLI

LocalAgentCLI is a local-first AI command-line assistant with a persistent shell, chat and agent modes, remote and local model backends, tool execution, session restore, and a centralized safety layer.

Features

• Interactive localagentcli shell with slash commands, history, tab completion, and streaming output
• Remote provider support for OpenAI-compatible, Anthropic, and generic REST APIs
• Local model support for MLX, GGUF, and safetensors backends
• Chat mode with context compaction, pinned instructions, and automatic repository-root AGENTS.md loading
• Agent mode with planning, tool execution, approvals, rollback, and undo support
• Persistent config, model registry, sessions, logs, and cache under ~/.localagent/

Installation

Recommended:

pipx install localagentcli

Development install:

git clone https://github.com/rainzhang05/LocalAgentCLI.git
cd LocalAgentCLI
pip install -e ".[dev]"

Optional backend extras can be installed manually:

pip install "localagentcli[mlx]"
pip install "localagentcli[gguf]"
pip install "localagentcli[torch]"
pip install "localagentcli[all]"

LocalAgentCLI also prompts to install missing backend dependencies automatically the first time you load a local model that needs them.

Quick Start

Launch the shell:

localagentcli

localagent remains available as a compatibility alias.

On first launch the setup wizard creates ~/.localagent/config.toml.

Common commands:

/help
/status
/setup
/mode chat
/mode agent
/set
/models
/providers list
/session save my-work
/session load my-work

Local and Remote Models

Remote providers:

• /providers add
• /providers list
• /providers test <name>

Local models:

• /set to choose the active local model or provider model interactively
• /models for the interactive Hugging Face picker (backend → family → live Hub-discovered model) with continuously refreshed download progress
• /models install hf <repo>
• /models install url <url>
• /models list
• /models inspect <name[@version]>

Development

Run the required checks locally:

python -m pytest --cov=localagentcli --cov-fail-under=80
ruff check .
ruff format --check .
mypy localagentcli/
python -m build
python -m twine check dist/*

Documentation

Project documentation lives in the repository:

• Architecture
• Current State
• Roadmap
• Packaging and Release