QwenPaw v2 Personal AI Assistant

QwenPaw v2.2.2 is an open-source personal AI assistant from the AgentScope team. This template runs it around the clock with a web console, chat apps such as Telegram, Discord and Slack, long-term memory on a volume, and a login created before the app starts.

What you get

  • QwenPaw v2.2.2 from the official image, with web console, REST API, agents, skills, cron and memory
  • One Railway volume at /data for chats, memory and encrypted keys
  • Login from generated QWENPAW_AUTH_USERNAME and QWENPAW_AUTH_PASSWORD; the service will not start without one
  • Chat apps (Telegram, Discord, Slack, DingTalk, Feishu, QQ, WeCom) enabled from variables, with sender approval
  • Tool Guard forced on so risky tool calls wait for your approval; telemetry off by default

Good for

  • A private assistant you message from Telegram, Discord, Slack or Feishu
  • Scheduled briefings, reminders and recurring checks delivered to a chat
  • Research and document work with a real Chromium browser
  • A REST API that streams agent replies into your own scripts

Cost

The app alone measured 540 MB RSS idle in a native run (not on Railway); the template's sizing note puts the full container at roughly 0.7 to 0.8 GB idle, more while Chromium runs.

Services in this template

  • QwenPaw · from baranberkay96/qwenpaw-railway

Questions people ask

How do I log in?

Open the public URL once healthy and sign in with QWENPAW_AUTH_USERNAME and QWENPAW_AUTH_PASSWORD from the Variables tab. Then add a model provider key under Settings > Models. The login cannot be switched off in this template.

Do I need a model key?

Not to deploy. You need an API key for at least one supported provider, entered in the console after login.

What are the security caveats?

The agent's shell tool and the console terminal run as root inside the container with access to the volume and network, and Railway offers no kernel sandbox for them. Set QWENPAW_TOOL_GUARD_DENIED_TOOLS=execute_shell_command to remove shell access.

Can I run more than one replica?

No. QwenPaw is a single-writer process with file-based state, so keep one replica and scale vertically.

How do I upgrade?

Change the pinned tag and digest in the Dockerfile to a newer release and redeploy; everything on the volume is kept. Download a backup from the console first.

Full template README

Deploy and Host QwenPaw with Railway

Run QwenPaw, the open-source personal AI assistant from the AgentScope team, around the clock on Railway. It answers in its own web console and in Telegram, Discord, Slack, DingTalk, Feishu, QQ and WeCom, keeps long-term memory on a persistent volume, and runs skills and scheduled tasks while you are away.

About Hosting QwenPaw

QwenPaw is a single Python service: a web console and REST API, agents with tools (shell, files, a real Chromium), skills, cron jobs, heartbeats and long-term memory. Self-hosting it needs a machine that never sleeps, durable storage for chats, memory and encrypted keys, and a login, because upstream ships with authentication off. This community template runs the official agentscope/qwenpaw:v2.2.2 image with all state on one Railway volume, creates the login from generated credentials before the app starts and refuses to start without one, switches chat apps on from variables with sender approval, and sends logs to Railway.

Common Use Cases

  • A private assistant you message from Telegram, Discord, Slack, DingTalk or Feishu that remembers your preferences across conversations
  • Scheduled briefings, reminders and recurring checks delivered to a chat, set up in plain language
  • Research and document work: web browsing with a real Chromium, PDF and Office files, summaries saved to the workspace
  • Several agents with their own personas, models and tools in one instance, reachable from phone and desktop
  • A REST API that streams agent replies into your own scripts and services

Dependencies for QwenPaw Hosting

  • An API key for at least one model provider (DashScope, OpenAI, Anthropic, Gemini, OpenRouter, DeepSeek or any OpenAI- or Anthropic-compatible endpoint), entered in the console after the first login. Not needed to deploy.
  • Optional bot credentials for the chat apps you want to use
  • A Railway volume at /data (included)

Deployment Dependencies

Implementation Details

Service Source Purpose
QwenPaw Dockerfile in services/qwenpaw on agentscope/qwenpaw:v2.2.2 (pinned by digest) Console, API, agents, tools, chat channels, cron and memory on port 8088; healthcheck /api/version; volume at /data; restart policy always

First login

  1. Open the service's public URL once the deployment is healthy.
  2. Sign in with QWENPAW_AUTH_USERNAME and QWENPAW_AUTH_PASSWORD from the service's Variables tab. Both are generated at deploy; the username is random so nobody can lock you out by guessing it.
  3. Go to Settings → Models, add your provider key and pick the default model.
  4. Say hello in Chat. The agent first asks a few questions to set up its persona.
  5. Chat apps: set TELEGRAM_BOT_TOKEN, DISCORD_BOT_TOKEN, SLACK_BOT_TOKEN + SLACK_APP_TOKEN, DINGTALK_CLIENT_ID + DINGTALK_CLIENT_SECRET, FEISHU_APP_ID + FEISHU_APP_SECRET, QQ_APP_ID + QQ_CLIENT_SECRET or WECOM_BOT_ID + WECOM_SECRET and redeploy, or use Control → Channels. Message your bot, then approve yourself under Pending approvals: unknown senders cannot reach the agent until you approve them.

Login: change QWENPAW_AUTH_PASSWORD (or the username) and redeploy to reset it; all sessions are signed out. A password changed in the console is kept until you change the variable. The login cannot be switched off in this template.

Language and timezone: QWENPAW_AGENT_LANGUAGE (en, zh, ru or id) and TZ are applied on the first boot; later, change them in the console.

Security notes: the agent's shell tool and the console's terminal run as root inside the container with access to the volume and network; Railway offers no kernel sandbox for them. Tool Guard is forced on, so risky tool calls wait for your approval. Set QWENPAW_TOOL_GUARD_DENIED_TOOLS=execute_shell_command to remove shell access. Upstream's usage telemetry is off by default (QWENPAW_TELEMETRY_DISABLED=true).

Scaling: QwenPaw is a single-writer process with file-based state. Keep one replica and scale vertically. The template is sized for the Hobby plan (about 0.7-0.8 GB RAM idle, more while the browser runs).

Pinning and upgrades: the Dockerfile pins agentscope/qwenpaw:v2.2.2 by tag and digest. To upgrade, change both to a newer release from https://github.com/agentscope-ai/QwenPaw/releases and redeploy; everything on the volume is kept. Download a backup from the console first.

Why Deploy QwenPaw on Railway?

An assistant that lives in your chat apps is only useful if it is always online and never forgets. Railway keeps QwenPaw running with automatic restarts, a persistent volume, HTTPS, logs and SSH access, without a server to maintain. You pay for the RAM and CPU the assistant actually uses; model usage is billed by your provider.

View QwenPaw v2 Personal AI Assistant on railway.com

Similar templates

TemplateCategoryDeploymentsDeploys
9Router
Self-hosted AI router for Claude, Gemini, Cursor & OpenAI tools.
AI/ML9,734
flowise-railway
Flowise - low-code LLM apps builder
AI/ML8,512
LibreChat
Open-source ChatGPT clone: multi-model, agents, search, multi-user auth
AI/ML6,762
Hermes Agent by Nous Research
Template for Hermes Agent by Nous Research
AI/ML6,413
OpenClaw
[Homebrew support] The AI that actually does things
AI/ML5,167
Hermes Agent
Self Host Hermes Agent. Autonomous AI agent w/ self-improving, tools & more
AI/ML4,092