jackwener--opencli/docs/adapters/browser/qwen.md

Qwen (通义千问)

Drive Qianwen / Qwen chat from the terminal. All commands run through your existing browser session — no API key needed.

Mode: 🔐 Browser · Domain: www.qianwen.com

Commands

Command	Description	Access
opencli qwen status	Page availability, login state, current model and session	read
opencli qwen history	List recent conversations (requires login)	read
opencli qwen read	Read messages in the current conversation	read
opencli qwen detail <id>	Open a conversation by ID and read its messages	read
opencli qwen ask <prompt>	Send a prompt and wait for the assistant reply	write
opencli qwen send <prompt>	Fire-and-forget: send a prompt without waiting	write
opencli qwen new	Start a fresh conversation	write
opencli qwen image <prompt>	Generate images (AI 生图) and save them locally	write

Usage Examples

# Sanity check
opencli qwen status

# Recent conversations
opencli qwen history --limit 10

# Read the active conversation as markdown
opencli qwen read --markdown true

# Read a specific historical conversation by ID (or full URL)
opencli qwen detail abcd1234ef567890abcd1234ef567890
opencli qwen detail https://www.qianwen.com/chat/abcd1234ef567890abcd1234ef567890 --markdown true

# Ask a question and wait for the reply
opencli qwen ask "用一段话解释 transformer attention"

# Ask in a brand-new chat with DeepThink turned on
opencli qwen ask "推导 KL 散度的链式法则" --new true --think true

# Fire-and-forget (don't wait for the reply)
opencli qwen send "继续上面的推导" --think true

# Start a new conversation
opencli qwen new

# Generate an image and save it locally
opencli qwen image "a watercolor fox sketching on paper"

# Custom output directory
opencli qwen image "cyberpunk skyline" --op ~/Downloads/qwen-images

# Skip download — just print the share link
opencli qwen image "a tiny robot" --sd true

Options

ask / send

Option	Description
prompt	Prompt to send (required positional)
--new	Start a new chat before sending (default: false)
--think	Enable 深度思考 (DeepThink)
--research	Enable 深度研究 (DeepResearch)
--markdown	(ask only) Emit assistant reply as markdown
--timeout	(ask only) Max seconds to wait for the reply (default: 120)

image

Option	Description
prompt	Image prompt (required positional)
--op	Output directory (default: ~/Pictures/qianwen)
--new	Start a new chat before generating (default: true)
--sd	Skip download; only print the conversation link
--timeout	Max seconds to wait for the image response (default: 180)

read

Option	Description
--markdown	Emit assistant replies as markdown (default: false)

detail

Option	Description
id	Session ID (32-char hex) or full https://www.qianwen.com/chat/<id> URL (required positional)
--markdown	Emit assistant replies as markdown (default: false)

history

Option	Description
--limit	Max conversations to list (default: 20, max 100)

Output Columns

Command	Columns
status	Status, Login, Model, SessionId, Url
history	Index, Title, Updated, Url
read	Role, Text
detail	Role, Text
ask	Role, Text
send	Status, Prompt
new	Status
image	Status, File, Link

Prerequisites

• Chrome is running
• You are already signed into www.qianwen.com
• Browser Bridge extension is installed

Notes

• read works without login (guest mode), but history and most write flows require an authenticated session
• Qwen commands default to persistent site sessions, so consecutive qwen ask / qwen read / qwen image invocations continue in the same Qwen page. Pass --site-session ephemeral for a one-shot tab.
• ask waits for the streaming reply to finish; send returns immediately after submission
• DeepThink (--think) and DeepResearch (--research) toggle the corresponding composer chips before submitting
• Generated image files are timestamped to avoid overwriting prior runs
• status returns Model / SessionId as null when they cannot be detected (e.g. guest mode, page still loading) rather than a string sentinel — branch on null in agent code
• DOM or product changes on Qwen can break composer detection — opencli qwen status is the quickest sanity check
• limit is validated and rejected with ArgumentError if non-positive or above the documented max (e.g. history max 100); no silent clamp