
Deploy Tarsee
24/7 AI agent on Claude, ChatGPT or any API key. Web, chat apps, voice.
Tarsee
Just deployed
/data
Deploy and Host Tarsee on Railway
Tarsee is your own AI agent, running 24/7, that you talk to from anywhere: web, Telegram, Discord, WhatsApp, Slack, email or voice. It remembers everything, runs tools and schedules tasks. Run it on your Claude or ChatGPT plan, on your own API key (Anthropic, OpenAI, OpenRouter or DeepSeek), or free on a model you run yourself. It's a lightweight Node.js server with no build step.
Why Deploy
- Your own AI agent running 24/7, no laptop needed
- Choose your AI: Claude plan, ChatGPT plan, Anthropic, OpenAI, OpenRouter, DeepSeek or Local AI (Ollama, LM Studio, llama.cpp). Switch any time, even for a single chat
- Talk to it from web, Telegram, Discord, WhatsApp, Slack, email or voice, all with the same memory and tools
- Persistent memory with deep search across sessions
- Browser automation with a stealth browser and captcha solving
- Scheduled tasks, daily briefings and cron jobs that run while you sleep
- Voice with local speech-to-text and a local voice, no voice API key needed
- Cost controls: a monthly spending cap, a backup AI, and a Usage page that shows what each AI cost
- REST API for iOS Shortcuts, scripts and automations
About Hosting Tarsee
Tarsee deploys as a single Docker container on Railway with a persistent volume at /data. The only variable you need to change is your 4-digit PIN (SETUP_PASSWORD); the encryption key and Node env are pre-configured. Open the URL, log in with your PIN, and the setup wizard asks you to choose your AI: sign in once with your Claude or ChatGPT plan, or paste an API key, click Test, then Use this AI. Prefer variables? Set TARSEE_AI_PROVIDER plus that AI's key in Railway and the wizard is skipped. Sign-ins are kept on the volume and keys are stored encrypted, so both survive restarts and redeploys. Tarsee needs about 500 MB of RAM. Voice works right away: on first boot it downloads its local voice (Phonon, about 72 MB) to the volume, and Edge TTS speaks in the meantime. To get new features, connect your service to projectservan8n/Tarsee under Settings > Source: every push to main then deploys, and your data on /data is kept.
Common Use Cases
- Personal AI assistant on any device: phone, laptop, tablet
- An agent for a team or customers on API keys, with "Only use API keys" locked on
- Voice agent with local speech-to-text and text-to-speech
- Browser automation: forms, screenshots, research, captchas (with a 2Captcha or Capsolver key)
- Scheduled tasks, daily briefings and cron jobs
- One AI on web, Telegram, Discord, WhatsApp, Slack and email at once
- A free, private agent on your own model server, run as another Railway service on the private network
Dependencies for Tarsee Hosting
- One AI account: a Claude plan (Pro, Max or a Team seat; Max lasts much longer), a ChatGPT plan (Free works, with limits), an API key from Anthropic, OpenAI, OpenRouter or DeepSeek, or your own Ollama, LM Studio or llama.cpp server
- Railway account with a volume mounted at
/data - Telegram bot token (optional, from @BotFather)
- Discord bot token (optional, from the Developer Portal)
- Slack app tokens (optional, Socket Mode), WHAPI Cloud for WhatsApp (optional), any IMAP/SMTP mailbox for email (optional)
Deployment Dependencies
- Docker (handled by Railway automatically)
- Node.js 22 (included in the Docker image)
- Python 3 (included, for faster-whisper speech-to-text and the local voice)
- Playwright + Chromium (included, for browser automation)
- Claude Code CLI and OpenAI Codex (included, at pinned versions)
Template Content
Tarsee
projectservan8n/Tarsee
