You're almost live.
Three steps and your agent is online.
setup.sh handles most of this automatically.
When you run the setup script from your kit, it installs everything you need and walks you through each step. The environment guides are here if you need to install something manually, troubleshoot a specific step, or set up a second machine.
Unzip the kit
Extract the downloaded ZIP file to a permanent folder on your computer.
Run the setup script
Open your terminal, navigate to the extracted folder, and run the one-command setup.
chmod +x setup.sh && ./setup.sh
Enter your API key
When prompted, select your model provider (OpenAI, Anthropic, or Google) and paste your API key. The script will securely configure your local environment and bring your agent online.
What happens after setup?
Your agent is now running locally. Open Telegram or the local web UI and simply say "Hello".
The agent will introduce itself, ask a few conversational questions to understand your preferences, and automatically schedule your first daily briefing. No complicated dashboards or configuration required.
Talking to Your Team
You can talk directly to any agent in your kit, not just your main agent. Ask your main agent to "put me through to Snag" and they’ll hand off the conversation, or start a direct session when you want a specialist to stay focused on one task.
- Route through your main agent when that’s easiest, like asking to talk to Snag or Rosetta.
- Go direct to the supporting agent that fits the job, whether you need troubleshooting, explanation, build work, content, or research.
- Use Discord thread mode for persistent specialized sessions with full context.
Working with AI agents — what to expect
Your agents are powerful, but they're not perfect — and that's by design. AI models are always learning. Context doesn't always carry perfectly between sessions. A task that ran cleanly last week might need a little more detail next time.
Think of your agent like a great employee, not a perfect robot. They get better the more context you give them, they'll tell you when they're unsure, and they iterate well when you say what was off.
- ✓ Keep MEMORY.md updated with things your agent should always know
- ✓ Be specific — "write a short email declining this meeting" beats "handle this"
- ✓ If a task came back wrong, add context and ask for a revision
- ✓ If something keeps going wrong, ask your agent to loop in Snag or Rosetta
The goal isn't perfection — it's compound improvement. Every good interaction makes the next one better.
FAQ
What if I don't have Node.js installed?
The script will pause and prompt you to install it. Just grab the latest LTS version from nodejs.org, install it, and run setup again.
Can I switch agents or models later?
Absolutely. Your config.json is fully editable, and you can swap to a different agent persona just by telling your current agent to pull a different profile from The Library.
What does it cost to run?
You only pay for what you use directly to the AI provider. Typical usage runs $20-50/month depending on how much automation you have running and which model you choose.