a1ai-cli

A1Ai in your terminal — seven specialist AI agents for code, design, art direction, climate intelligence and XR.

powered by A1A®

npm install -g a1ai-cli
a1ai "explain this stack trace"

The package is a1ai-cli; the command is a1ai. npm's typosquat check strips punctuation, which makes the bare name a1ai indistinguishable from the unrelated a1-ai already on the registry — so the package took the longer name and the binary kept the short one.

No dependencies. The whole CLI is a single bundled file, so a global install pulls nothing else.

The agents

| | Agent | Product | Role | |---|---|---|---| | 🧠 | neural | A1Ai | General assistant | | ⚔️ | codex | Call Of Codes | Coding tutor and hackathon judge | | 🎨 | ambara | Ambara Art | Mockup and design generation | | 🎬 | kata | Kata Dimension | Creative director | | 🛰️ | overwatch | Overwatch Cloud | Climate, conservation, astronomy | | 🌐 | jarvis | CyberChat City | XR avatar assistant | | ✨ | muse | A1Ai for artists | Craft and research partner |

They are agents, not personas: each carries its own tools — live weather and air quality, web search, a safe expression evaluator, colour palettes with WCAG contrast checks, rubric scoring, and (locally) file reading and code execution.

Usage

a1ai                                   # interactive session
a1ai "what is the CAP theorem?"        # one-shot
a1ai -a overwatch "rain in Kigali?"    # pick an agent
cat error.log | a1ai "what broke?"     # pipe context in
git diff | a1ai -a codex "review this"

Inside a session: /agent, /model, /clear, /help, /exit. Ctrl-C cancels the current answer without killing the session.

a1ai login          # sign in with your A1A ID account
a1ai agents         # list agents
a1ai models         # models your tier can reach
a1ai keys new ci    # mint an API key
a1ai doctor         # diagnose setup problems

Two ways to run

Signed in (default) — talks to an A1Ai server, which handles quotas, billing and history:

a1ai login

Direct — when you are not signed in and a provider key is in the environment, it talks straight to OpenRouter or NVIDIA with your own key:

export OPENROUTER_API_KEY=sk-or-v1-...
a1ai "hello"

Direct mode also enables local tools (read_file, write_file, list_files, run_code) scoped to the directory you ran a1ai from. The hosted server never registers those — it is shared infrastructure, your machine is not.

run_code is not a security sandbox. It runs real JavaScript or Python in a child process with a stripped environment, a timeout and capped output — enough to stop a runaway loop, not enough to contain hostile code.

Environment

| Variable | Purpose | |---|---| | A1AI_BASE_URL | server to use (default https://a1ai-api.onrender.com) | | A1AI_API_KEY | authenticate without a1ai login — for CI | | OPENROUTER_API_KEY | direct provider mode | | NVIDIA_API_KEY | direct provider mode, premium models | | NO_COLOR | disable colour |

Colour and spinners switch off automatically when output is piped.

Tiers

Freemium — free OpenRouter models, 60 requests a day. Premium ($6.99 / 30 days) — adds NVIDIA NIM models, unmetered requests, longer answers and deeper tool chains. a1ai upgrade walks you through it.

Part of the A1A ecosystem — Call Of Codes · Ambara Art · Kata Dimension · Overwatch Cloud · CyberChat City.