bardgent 1.0.1

A command-line coding agent (OpenAI-compatible API; default NVIDIA Nemotron)

Bardgent

A terminal coding agent with tools for files, shell, web, memory, skills, sub-agents, and scheduled tasks.

PyPI: https://pypi.org/project/bardgent/

---

Install

Requires Python 3.8+.

pip install bardgent

Or with pipx (recommended for CLI apps — isolated env + command on PATH):

pipx install bardgent

Upgrade later:

pip install --upgrade bardgent
# or
pipx upgrade bardgent

---

Configure API key

The env var name is historical (GEMINI_API_KEY). Any OpenAI-compatible key works with the configured base URL (default: NVIDIA).

Recommended: config file

Create a .env file in the Bardgent config directory.

macOS / Linux

mkdir -p ~/.bardgent
nano ~/.bardgent/.env

Windows (PowerShell)

New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.bardgent"
notepad "$env:USERPROFILE\.bardgent\.env"

Put this line in the file (no spaces around =):

GEMINI_API_KEY=your_key_here

Optional:

TELEGRAM_BOT_TOKEN=optional_telegram_bot_token

OS	Config directory	Env file
macOS / Linux	~/.bardgent/	~/.bardgent/.env
Windows	%USERPROFILE%\.bardgent\	%USERPROFILE%\.bardgent\.env

Alternative: environment variable

macOS / Linux

export GEMINI_API_KEY=your_key_here

Windows (PowerShell)

$env:GEMINI_API_KEY = "your_key_here"

---

Run

bardgent

If the command is not found after pip install, use:

python -m bardgent

On Windows, if python is not found, try:

py -m bardgent

---

Defaults

Setting	Value
API base	https://integrate.api.nvidia.com/v1
Model	nvidia/nemotron-3-ultra-550b-a55b
Config dir	~/.bardgent/ (Windows: %USERPROFILE%\.bardgent\)

---

Modes

Mode	Behavior
normal	Approve Write/Edit/Bash (and similar) per action
auto	Auto-approve non-dangerous actions; dangerous shell still prompts
plan	Read-only tools only; agent proposes a plan, then you choose how to execute

Cycle with Shift+Tab or /normal, /auto, /plan.

---

Tools (high level)

• Files: Read, Write, Edit, Undo, Glob, Grep
• Shell: Bash (cwd persists), background jobs via ListJobs / Await
• Web: WebSearch (DuckDuckGo), Fetch
• Memory: save / list / delete long-term notes (~/.bardgent/Bardgent.md)
• Skills: Claude Code–style SKILL.md packs; /skills, /skill install <github_url>
• Delegation: Task / Tasks (parallel sub-agents)
• Schedule: recurring or one-off tasks; /schedule, /schedules. A detached scheduler daemon keeps firing tasks after you close the terminal (auto-started with Bardgent). Manage with /schedule daemon status|start|stop. Log: ~/.bardgent/scheduler.log.

---

Project instructions

Place any of these in the project root (all that exist are loaded):

• AGENTS.md
• .bardgent/AGENTS.md, .bardgent/instructions.md, .bardgent/RULES.md

---

Skills locations

1. ./.bardgent/skills/<name>/SKILL.md (project)
2. ~/.bardgent/skills/<name>/SKILL.md (user)
3. Package skills/ (optional bundled defaults)

---

Permissions

Per-project file: .bardgent/permissions.json

{
  "auto_approve_bash_prefixes": ["git status", "pytest"],
  "auto_approve_tools": ["Fetch", "Write", "Edit"],
  "extra_dangerous_patterns": []
}

---

Useful commands

Type /help in the REPL for the full list. Highlights: /resume, /checkpoints, /restore, /telegram, /summary, /clear.

---

Develop from source

git clone https://github.com/bardiaegz/BardGent.git
cd BardGent
python -m venv venv

# macOS / Linux
source venv/bin/activate

# Windows (PowerShell)
# .\venv\Scripts\Activate.ps1

pip install -e .
bardgent

---

License

MIT — use and modify as you like for personal / project use.