Moltbootee

Moltbootee is a standalone Node.js CLI for authenticating with Qwen or Gemini and sending coding prompts from the same terminal UX.

Why it differs from Qwen-portal-OAUTH.md

The markdown spec in this workspace describes an older OAuth authorization-code flow with a localhost callback such as http://127.0.0.1:7777/oauth/callback.

When I verified the current live implementation on March 7, 2026, Qwen Code was using a different flow:

• Device-code OAuth at https://chat.qwen.ai/api/v1/oauth2/device/code
• Token polling at https://chat.qwen.ai/api/v1/oauth2/token
• Browser authorization at https://chat.qwen.ai/authorize?...
• Model access through an OpenAI-compatible endpoint, typically https://dashscope.aliyuncs.com/compatible-mode/v1

Moltbootee follows the current live flow so the CLI is actually usable.

Fast start

node ./bin/molty.js

If molty is linked onto your PATH, typing just molty starts the interactive chat shell immediately.

molty also supports multiple stored accounts. In the interactive shell, Tab then A cycles to the next stored account.

If Gemini gemini-3-pro-preview is temporarily out of capacity, Moltbootee automatically retries with the next built-in Gemini model instead of dropping the whole shell.

Providers

• qwen: Qwen OAuth device-code login plus coder-model
• gemini: Google OAuth localhost callback plus Gemini CLI-style Code Assist models such as gemini-3-pro-preview

Examples:

moltbootee login
moltbootee login gemini
moltbootee login qwen
moltbootee login --provider gemini
moltbootee chat --model gemini-3-pro-preview "Explain this stack trace"
moltbootee models --provider gemini

Commands

node ./bin/moltbootee.js help
node ./bin/moltbootee.js login
node ./bin/moltbootee.js login gemini
node ./bin/moltbootee.js login --provider gemini
node ./bin/moltbootee.js login --account work
node ./bin/moltbootee.js accounts
node ./bin/moltbootee.js account use work
node ./bin/molty.js
node ./bin/moltbootee.js models
node ./bin/moltbootee.js chat "Explain this bug"
node ./bin/moltbootee.js skill debug "Find the bug in this stack trace"
node ./bin/moltbootee.js agent reviewer "Review this function"
node ./bin/moltbootee.js skills
node ./bin/moltbootee.js agents
node ./bin/moltbootee.js doctor
node ./bin/moltbootee.js logout

If you want the command on your PATH, run:

npm.cmd link

Then use:

moltbootee login
moltbootee login gemini
moltbootee login --provider gemini
molty

npm.cmd link now exposes both commands:

• moltbootee: full command surface
• moltybootee: alias for moltbootee
• molty: chat-first shortcut

Multiple accounts

Accounts are stored in the same OAuth file and one account is marked active.

Examples:

moltbootee login --account personal
moltbootee login --provider gemini --account google
moltbootee login --account work
moltbootee accounts
moltbootee account use work
moltbootee chat --account personal "review this script"

Inside molty:

• /login gemini logs into Gemini and switches the session
• /login qwen logs into Qwen and switches the session
• /logout removes the current account session
• /accounts lists stored accounts
• /account shows the current one
• /account use work switches directly
• /provider gemini switches the session provider
• Tab then A cycles to the next stored account

Built-in skills

Moltbootee now ships with built-in reusable task skills:

• build
• debug
• review
• refactor
• test
• docs
• shell
• security
• performance
• api
• ui
• git
• sql
• explain

Use them like:

moltbootee skill debug "Why is this promise never resolving?"
moltbootee chat --skill review "Review this Express middleware"

Built-in agents

Moltbootee also includes agent personas:

• coder
• architect
• reviewer
• debugger
• test-engineer
• devops
• security
• frontend
• backend
• data
• teacher
• product

Use them like:

moltbootee agent architect "Design a plugin system"
moltbootee chat --agent security "Threat model this OAuth callback flow"

Interactive shell

Inside molty, use slash commands:

• /help
• /accounts
• /account [id]
• /login [provider]
• /logout [id]
• /provider [id]
• /skills
• /agents
• /use-skill <id>
• /use-agent <id>
• /model <id>
• /model qwen
• /model gemini
• /model flash
• /stream on|off
• /system <text>
• /status
• /doctor
• /history
• /save [file.md]
• /clear
• /retry
• /models
• /exit

Storage

OAuth credentials are stored at:

~/.moltbootee/oauth.json

Example structure:

{
  "version": 3,
  "active_account_id": "work",
  "accounts": {
    "work": {
      "account_id": "work",
      "label": "work",
      "provider": "qwen",
      "access_token": "....",
      "refresh_token": "....",
      "token_type": "Bearer",
      "expires_at": 1772861575000,
      "resource_url": "https://dashscope.aliyuncs.com/compatible-mode",
      "created_at": "2026-03-07T10:00:00.000Z",
      "updated_at": "2026-03-07T10:00:00.000Z"
    }
  }
}

Request format

Qwen uses the OpenAI-compatible POST /chat/completions shape:

{
  "model": "coder-model",
  "messages": [
    {
      "role": "user",
      "content": "Write a tokenizer in Rust"
    }
  ],
  "stream": true,
  "temperature": 0.2
}

The CLI sends:

• Authorization: Bearer <access_token>
• Content-Type: application/json

Gemini uses the same Google OAuth and Code Assist backend shape as the official Gemini CLI, including generateContent and streamGenerateContent against https://cloudcode-pa.googleapis.com/v1internal.

Notes

• The model alias qwen-portal/coder-model is accepted and normalized to coder-model.
• gemini-3-pro-preview, gemini-3-flash-preview, gemini-2.5-pro, gemini-2.5-flash, and gemini-2.5-flash-lite are built in as official Gemini CLI model ids.
• moltbootee login gemini and moltbootee login qwen are shorthand for provider-specific login.
• In molty, /model qwen and /model gemini switch provider and default model together.
• If the access token is near expiry, Moltbootee refreshes it automatically.
• If refresh fails with an invalid refresh token, the CLI clears the local token file and requires a new login.
• molty with no arguments drops directly into the prettier interactive shell.