---
name: quiq-login
description: Operate a Quiq customer-experience account on the user's behalf via Quiq's AI Connector CLI (@quiq/cli). Use when the user wants to work in their Quiq account — e.g. pull conversations, reporting, or account configuration. Installs the connector if needed and connects.
---

# Quiq AI Connector

Quiq's AI Connector is a command-line tool (`@quiq/cli`) you run to operate a Quiq
customer-experience account on the user's behalf. Get an approved session going; the
connector then tells you how to drive it.

## 1. Install — only if it isn't already there

On some setups the connector persists between conversations, so check first:

    quiq --version

If that fails, install it:

    npm install -g @quiq/cli

If a command later reports the CLI is out of date: `npm install -g @quiq/cli@latest`.

## 2. Connect (the one step the user does)

1. Ask for the user's Quiq web address if you don't have it (e.g. `theircompany.goquiq.com`).
2. Start the connection, setting `--client` to the assistant you are:
   - Claude Code → `claude-code`
   - Claude Desktop → `claude-desktop`
   - Claude Cowork → `claude-cowork`
   - Claude.ai → `claude.ai`
   - Codex → `codex`
   - ChatGPT → `chatgpt.com`
   - GitHub Copilot → `github-copilot`
   - Anything else → `other`

       quiq session start --host <their-domain> --client <your-client>

3. It prints an approval URL. Hand it to the user as a Markdown link:

   `[Approve Quiq access](<the-url>)`

   Tell them to open it, sign in, and click Approve. There's nothing to paste back.
4. Once they confirm they've approved:

       quiq session complete

5. Pass `--session <id>` on every command afterward; one session per conversation. If a later
   command says the session needs re-approval, repeat steps 2–4.

## 3. Bootstrap, then work

    quiq bootstrap --session <id>

This materializes the live surface for your session — the capabilities you can call, task
recipes, and reference help — to disk, and prints how to drive it. Follow what it prints.

## Notes

- Lead with outcomes in the user's terms, not raw command names.
- Everything is scoped to what the user approved.