LinkAI CLI

The command-line interface for the LinkAI agent platform.

English · 中文

LinkAI CLI brings the full LinkAI platform to the terminal for both agents and people — models, apps, knowledge bases, databases, workflows, plugins, and accounts. It gives AI agents a rich set of platform capabilities as infrastructure, extending what they can do.

Capabilities

The CLI exposes two kinds of capabilities: platform resources (manage and use the resources you've configured on LinkAI) and model capabilities (call models to generate content).

📦 Platform resources

Resource	Command	Description
Apps	app	List, inspect, create, update, and delete AI apps
Knowledge	knowledge	Vector-search private knowledge bases; import files and manage bases
Database	database	Query databases and schemas, run SQL, create builtin databases and tables
Workflow	workflow	Run workflows orchestrated on LinkAI; create, update, and delete them
Plugin	plugin	List, inspect, and run platform plugins
Account	account	Account profile, credits, and recharge

🧠 Model capabilities

Capability	Command	Description
Chat	chat	Talk to an AI app or language model, with multi-turn sessions
Image	image	Text-to-image; returns an image URL
Video	video	Text-to-video, with built-in polling until the result is ready
Speech	audio	Text-to-speech (TTS), optionally saved to a file
Models	model	List available models (LLM / IMAGE / VIDEO)

Run linkai <command> --help for the full flags of any command.

Installation

Option 1 — npm

Requires Node.js 16 or later. npm installs the command to its global bin directory, which is already on your PATH:

npm i -g linkai-cli

Option 2 — let an agent install it

Send an agent this message; it will install the CLI, fetch the skill, and start using it:

Read https://cdn.link-ai.tech/cli/install.md and follow it to install the CLI and skill, then start using it.

Option 3 — install script (no dependencies)

# macOS / Linux
curl -fsSL https://cdn.link-ai.tech/cli/install.sh | sh
# Windows (PowerShell)
irm https://cdn.link-ai.tech/cli/install.ps1 | iex

The script downloads the pre-built binary, sets up your PATH, and installs the agent skill into common AI tool directories (Claude Code, Cursor, Codex, CowAgent, etc.).

Other install methods (Homebrew / Go / Binary / Source)

• Homebrew: brew install MinimalFuture/tap/linkai
• Go: go install github.com/MinimalFuture/linkai-cli@latest
• Binary: download from GitHub Releases, extract, and put it on your PATH.
• Source: git clone, then make build && make install.

Install script environment variables

Variable	Description	Default
LINKAI_VERSION	Version to install	latest
LINKAI_INSTALL_DIR	Binary install directory	Auto (prefers a directory already on PATH)
LINKAI_SOURCE	Download source: cdn / github	cdn (falls back to GitHub if unreachable)
LINKAI_NO_SKILL	Set to 1 to skip installing the skill	—

Getting started

Three steps to your first call:

linkai auth login                     # 1. Log in (authorize in the browser)
linkai app list                       # 2. List apps and grab an app_code
linkai chat "Hello" --app <app_code>   # 3. Chat with the app

Use auth status to check your session and auth logout to sign out.

Using with agents

Once installed, agents learn how to drive the CLI through its bundled skill — no extra setup required. The skill lives in skills/linkai-cli/.

A few notes when calling from an agent:

• Add --json for structured, easy-to-parse output.
• Preview write operations with --dry-run; add --force to skip confirmations (e.g. knowledge delete).
• Run linkai <command> --help for a command's full flags.

Command reference

Command	Description
auth login / logout / status	Log in / out / show login status
app list / detail / create / update / delete	List and manage apps
chat "<message>" --app <code>	Chat with an app; --session for multi-turn
knowledge list / files / search / create / import / delete	Query, import, and manage knowledge bases
database list / tables / describe / exec / create / create-table	Query, run SQL, and create builtin databases/tables
workflow list / run <code> --input "<text>" / create / update / delete	List, run, and manage workflows
plugin list / detail / exec <code>	List and run plugins
image gen "<prompt>"	Text-to-image
video gen "<prompt>"	Text-to-video (waits for completion)
audio speech "<text>" [--output a.mp3]	Text-to-speech, optionally downloaded
model list [--type LLM|IMAGE|VIDEO]	List available models
account info	Account profile and credit balance
account credits / recharge / orders	Credit packages, recharge, and orders

See linkai <command> --help for each command's full flags.

Permissions

Permissions are requested at login via --scope, in resource:action form. Read and content-generation scopes are granted by default; write scopes must be requested explicitly.

Scope	Description	Default
app:read	List and inspect apps	✅
app:create	Create apps	✅
user:read	Read user info	✅
chat:send	Chat with apps	✅
knowledge:read	Query knowledge bases	✅
knowledge:create	Create knowledge bases / add files	✅
db:read	Query databases / run SELECT	✅
image:gen / video:gen / audio:gen	Generate images / video / speech	✅
plugin:read / plugin:run	List / run plugins	✅
workflow:read / workflow:run	List / run workflows	✅
workflow:create	Create workflows	✅
score:read / score:buy	View / purchase credits	✅
app:update / app:delete	Update / delete apps	❌
knowledge:update / knowledge:delete	Update / delete knowledge bases	❌
workflow:update / workflow:delete	Update / delete workflows	❌
db:write	Database writes (INSERT/UPDATE/DELETE)	❌

Request extra scopes by logging in again, for example:

linkai auth login --scope "db:read db:write knowledge:update knowledge:delete"

More

Security

• Tokens: opaque tokens stored server-side and revocable; access lasts 2h, refresh lasts 7d, refreshed automatically before expiry; logout revokes them server-side.
• Device binding: every request carries X-Device-ID, and the server binds tokens to the device.
• Local storage: tokens go into the macOS Keychain (service linkai-cli), with a 0600 file fallback on other platforms; config lives in ~/.linkai/.
• Input / output hardening: dangerous Unicode is rejected, DDL is blocked in the database commands, and ANSI is stripped from table output.
• Resilience: 5xx responses are retried with exponential backoff; exit codes are structured (0 ok / 1 general / 2 validation / 3 auth / 4 network).

Shell completion

source <(linkai completion bash)                # Bash
source <(linkai completion zsh)                 # Zsh
linkai completion fish | source                 # Fish

Development

make build   # build (with version injection)
make test    # run tests
make lint    # golangci-lint

To add a command, follow the pattern in cmd/database/, register it in cmd/root.go, and declare its scope via permission.RequiredKey.

License

MIT