advai-cli

A unified CLI for local skills, browser automation, external CLIs, and terminal AI chat.

Why • Installation • Quick Start • Browser • Commands • TUI • Development

advai-cli is a unified command-line interface for managing local skills, driving a Chrome extension-backed browser bridge, working with external CLIs, and chatting with AI from the terminal through a single advai entrypoint.

It is designed as a lightweight Python-first core with npm and Homebrew distribution options, making it easy to install in different developer environments while keeping the runtime model simple, local-first, and predictable.

Why advai-cli

• One entrypoint for runtime inspection, local skill management, external CLI workflows, and terminal-native AI chat
• Chrome extension-backed browser automation through advai browser
• Python core for a small, dependency-light implementation
• npm wrapper for teams that prefer JavaScript-based distribution
• Homebrew formula for macOS-friendly installation
• OpenAI-compatible backend support for flexible AI provider integration
• Local-first state stored under ~/.advai

Highlights

• Inspect the current installation, runtime, and recommended update command with advai info and advai update
• Manage locally installed skills with install, list, info, update, and uninstall commands
• Sync installed skills into platform-specific agent directories such as Cursor, Claude Code, Codex, TRAE, and more
• Open, inspect, extract, and automate browser sessions with advai browser
• Discover and install supported third-party CLIs through the OpenCLI ecosystem
• Create local knowledge bases, add documents, search content, and resync from source files
• Proxy supported external CLIs through advai cli <name> ...
• Launch a terminal chat UI with configurable model, base URL, system prompt, and transcript export

Architecture

flowchart LR
    U[Developer] --> A[advai command]

    subgraph Distribution
        P[PyPI package]
        N[npm wrapper]
        B[Homebrew formula]
    end

    P --> A
    N --> A
    B --> A

    A --> C[Python CLI core]
    C --> S[Local Skills Manager]
    C --> BB[Browser Bridge]
    C --> O[OpenCLI Bridge]
    C --> T[Terminal TUI]
    C --> I[Runtime Inspection]

    S --> FS[~/.advai/skills]
    BB --> BSD[~/.advai/browser]
    BB --> EXT[Browser extension + local daemon]
    T --> API[OpenAI-compatible API]
    O --> OC[opencli binary]

Installation

Requirements

• Python 3.10+
• Node.js 14+ only if you install through npm
• A local Python runtime is still required when using the npm package
• opencli is required only for external CLI discovery, install, and passthrough execution
• A compatible browser extension/runtime is required only if you want to use advai browser

Install from PyPI

pip install advai-cli

Install from npm

npm install -g advai-cli

The npm package creates a private Python virtual environment during postinstall and installs the matching advai-cli version from PyPI into that environment.

Install from Homebrew

brew tap Advai-X/tap
brew install advai-cli

Install with the bootstrap script

./install.sh

Quick Start

1. Verify the installation

advai --help
advai info

2. Configure AI access

export ADVAI_API_KEY=your_api_key

3. Start the terminal chat UI

advai tui

4. Explore local skills and external CLIs

advai skill list
advai cli list

5. Verify browser automation

advai browser doctor
advai browser open demo https://example.com
advai browser exec demo "document.title"

Browser Automation

advai browser provides a Chrome extension-backed automation layer for opening pages, executing JavaScript, waiting for selectors, interacting with inputs, reading network events, and taking screenshots.

What it uses

• advai-cli provides the command surface, browser bridge client, and built-in local daemon
• A compatible browser extension/runtime provides the live browser connection
• Browser bridge state is stored under ~/.advai/browser
• The current release uses a fixed internal extension context for browser routing

Setup

1. Install the Chrome extension from the Chrome Web Store
2. Make sure the local browser runtime is available on this machine
3. Open the extension once in Chrome to complete the first-run setup
4. Sign in to any sites you want to automate in that browser context
5. Run advai browser doctor to confirm the daemon and extension are reachable

If the daemon is not already running, browser commands can auto-start the built-in advai browser daemon.

Core workflow

advai browser doctor
advai browser open demo https://example.com
advai browser open demo https://example.com --replace
advai browser wait demo --selector "h1"
advai browser exec demo "document.title"
advai browser type demo "#search" "advai browser"
advai browser click demo "button[type=submit]"
advai browser screenshot demo --output page.png

Common browser commands

Notes

• Browser sessions are named explicitly, such as demo, zhihu, or checkout-flow
• open creates a new tab by default; use open --replace to reuse the current tab for the session
• Commands wait for the active internal browser context before sending automation actions
• The browser bridge is local-first and does not require a remote automation service
• If doctor reports that browser support is missing, verify that the browser extension/runtime is installed and reachable locally

Common Commands

These are the canonical commands to use in docs, screenshots, and promotional assets:

advai info
advai browser doctor
advai browser open demo https://example.com
advai skill list
advai cli list
advai tui

TUI

The TUI connects to any OpenAI-compatible /chat/completions backend and runs entirely inside the terminal.

Useful examples

advai tui
advai tui --agent default
advai tui --model gpt-4o-mini
advai tui --base-url https://api.openai.com/v1
advai tui --system-prompt "You are a concise terminal coding assistant."
advai tui --timeout 180

Environment variables

advai-cli supports both ADVAI_* and part of the standard OPENAI_* naming for compatibility.

* ADVAI_API_KEY or OPENAI_API_KEY must be set before running advai tui.

In-session TUI commands

/help
/clear
/agent
/agent default
/model
/model gpt-4o-mini
/system You are a helpful assistant.
/save ./chat.md
/exit

Run /agent with no arguments to open an interactive agent picker. Run /model with no arguments to open an interactive model picker. Use the up and down arrow keys to choose a model, then press Enter to confirm.

Skills

Skills are stored locally under ~/.advai/skills.

advai skill list
advai skill search design --source skills_sh
advai skill platform list
advai skill info demo-skill
advai skill install https://github.com/your-org/skill-repo
advai skill install https://github.com/your-org/skill-repo --skill demo-skill
advai skill install https://github.com/your-org/skill-repo --skill demo-skill --platform cursor
advai skill install find-skills --source skills_sh
advai skill install skills.sh:vercel-labs/skills/find-skills
advai skill install https://skills.sh/vercel-labs/skills/find-skills
advai skill sync demo-skill --platform trae
advai skill sync demo-skill --platform omp_agent --project-dir /path/to/repo
advai skill unsync demo-skill --platform cursor
advai skill platform add custom_agent --name "Custom Agent" --path ~/.custom-agent/skills
advai skill platform override cursor --path ~/.cursor/skills
advai skill update demo-skill
advai skill uninstall demo-skill

Supported platforms

advai skill sync --platform <key> currently includes 51 built-in platform adapters, plus custom platform support through advai skill platform add.

Coding platforms

Lobster platforms

Custom platforms

• Use advai skill platform add <key> --name <name> --path <dir> to register any additional agent or internal platform that stores skills on disk.
• Use advai skill platform override <key> --path <dir> or --project-path <dir> to adapt built-in platform paths to your local setup.

Current scope:

• GitHub installs read from the repository root skills/ directory instead of copying the whole repo
• If the repo skills/ directory contains one skill, advai installs it automatically
• If the repo skills/ directory contains multiple skills, advai prompts whether to install all of them; use --skill <name> to choose one explicitly
• advai skill search <query> can discover skills from provider-backed sources such as skills.sh
• advai skill update <name> reuses the saved install source metadata, including provider-backed sources such as GitHub and skills.sh
• advai skill install supports GitHub repository URLs, provider-prefixed specs such as skills.sh:<owner>/<repo>/<skill>, and skills.sh skill page URLs
• Built-in platform adapters cover common coding and lobster-style agents, and custom platforms can be added locally
• advai skill sync supports symlink and copy modes for writing skills into platform-specific directories
• advai skill install <github-url> --platform <key> installs locally first, then syncs into each selected platform
• advai skill install <name> --source skills_sh searches skills.sh; interactive terminals can choose from multiple matches, while --yes requires an unambiguous result
• skills.sh search and detail lookups prefer the official API when ADVAI_SKILLS_SH_TOKEN (or SKILLS_SH_TOKEN / VERCEL_OIDC_TOKEN) is set, and fall back to public HTML pages otherwise
• skills.sh installs currently resolve to the underlying GitHub repository and selected skill slug before local installation
• advai skill uninstall <name> removes the local skill and cleans up any synced platform targets recorded in metadata
• Platforms with project-local skills directories can be targeted with --project-dir

External CLI Integration

External CLI support is powered by opencli.

advai cli list
advai cli info demo-cli
advai cli install https://github.com/your-org/cli-repo
advai cli install https://github.com/your-org/cli-repo --cli demo-cli
advai cli update demo-cli --yes
advai cli uninstall demo-cli --yes
advai cli demo-cli --help

Notes:

• advai cli list, advai cli info, advai cli install, advai cli update, and advai cli uninstall require opencli to be installed and available on PATH
• GitHub CLI installs read from the repository root clis/ directory and register each selected CLI via opencli external register
• If the repo clis/ directory contains multiple CLIs, advai prompts whether to install all of them; use --cli <name> to choose one explicitly
• Dynamic passthrough execution works only for CLIs exposed by the local OpenCLI registry
• advai-cli does not reimplement third-party CLIs; it provides a consistent entrypoint and installation surface

Knowledge Bases

Knowledge bases are stored locally under ~/.advai/kbs.

advai kb is also designed to be OKF-friendly. Based on the Open Knowledge Format (OKF) direction described by Google Cloud, the current implementation already works well with knowledge bundles that are organized as markdown files on disk, including files that contain YAML frontmatter.

advai kb create team-wiki
advai kb doc add team-wiki ./README.md
advai kb search team-wiki homebrew
advai kb sync team-wiki

Current scope:

• advai kb create <name> initializes a local knowledge base directory and metadata file
• advai kb doc add <name> <path> copies a source document into the knowledge base and tracks its original path
• advai kb search <name> <query> performs a simple case-insensitive text search over stored document content
• advai kb sync <name> refreshes stored copies from the original source paths and reports any missing files
• OKF-friendly ingestion: markdown-based knowledge docs can be added directly without conversion
• YAML frontmatter is preserved as plain text, so OKF metadata remains searchable and does not get stripped during add or sync
• This version does not yet implement full OKF-aware parsing, validation, or structured metadata querying; it focuses on filesystem-native storage and simple text retrieval first

Project Structure

advai/
  ai_client.py      OpenAI-compatible HTTP client
  browser_bridge.py Chrome extension bridge client
  cli.py            Main CLI entrypoint
  cli_manager.py    Install detection, update commands, OpenCLI integration
  kb.py             Local knowledge base storage and search helpers
  skill_platforms.py Platform adapter registry and path resolution
  skills.py         Local skill metadata management
  tui.py            Terminal chat UI
bin/
  advai.js          npm bridge that launches the private Python environment
  install-python.js npm postinstall script that creates the private virtualenv
docs/
  assets/
    hero.png        README hero image
Formula/
  advai-cli.rb      Homebrew formula
install.sh          Bootstrap installer

Development

Run from source

python -m venv .venv
source .venv/bin/activate
pip install -e .
advai --help
python -m advai.cli info

Packaging model

• PyPI ships the Python implementation directly
• npm creates a private virtual environment and installs the matching PyPI release into it
• Homebrew installs the Python package through a formula-managed virtual environment

Operational Notes

• Local state lives under ~/.advai
• Browser bridge logs and pid files are stored under ~/.advai/browser
• Knowledge bases are stored under ~/.advai/kbs
• Skills are stored under ~/.advai/skills
• The recommended self-update command changes depending on whether the tool was installed via pip, npm, or brew
• npm installation requires Python 3.10+ and prepares a private environment during postinstall

License

MIT