Skip to Content
FAQ

Frequently Asked Questions

Common questions from people setting up and using the agent system.

Getting Started

Do I need to know how to code?

No. The system is designed for non-technical users. The setup guide walks you through every step with exact instructions and expected results. You’ll use the terminal (a text-based interface on your Mac), but you don’t need to understand programming — just follow the steps.

What does this cost?

The main cost is an Anthropic API subscription for Claude. Usage is billed by how much you use it. Most users spend between $20-100/month depending on how many workflows they run. The software itself (Claude Code, the documentation, the templates) is free.

Does this work on Windows or Linux?

Currently, the automated scheduling features (nightly pipeline, morning brief generation) are Mac-only because they use macOS-specific automation. The core system (Claude Code, commands, agents) works on any platform, but you’ll need to set up scheduling differently. See the Developer Guide for platform-specific details.

How long does setup take?

About 2-3 hours for the basic setup. You can start using the system immediately after setup. Adding more connections and refining workflows happens over the following weeks as you discover what works best for you.

Using the System

Is my data private?

Your data stays on your computer and in your cloud accounts. The system accesses your email, calendar, and other tools using credentials you control. Conversations with Claude are processed by Anthropic’s API — see Anthropic’s privacy policy  for details on how they handle data.

Can it send emails on my behalf?

The system can draft emails for you to review and approve. It does not automatically send anything unless you explicitly set it up to do so. The default is always “draft and review” — you maintain full control.

What if it makes a mistake?

The system is designed with a “draft and approve” model for anything consequential. Email drafts, calendar changes, and important decisions are always presented for your review before action. For routine tasks (file organization, data collection), the system acts automatically but logs everything so you can review.

Can I undo something the system did?

Yes, in almost all cases:

  • Files are organized, not deleted — you can always find them
  • Email drafts require your approval before sending
  • Profile changes are saved in text files you can edit or revert
  • Everything is backed up nightly to your cloud storage

How do I make it better at drafting emails in my voice?

Two ways:

  1. Edit drafts and tell the system why — “That was too formal, make it more casual next time”
  2. Update your profile — Add examples of emails you like, describe your tone, note pet peeves

The more feedback you give, the faster it learns your style.

How It Works

Is this the same as ChatGPT?

Not exactly. ChatGPT is a general-purpose chatbot. This system is a structured, personalized assistant that:

  • Knows your preferences, constraints, and goals (from your profile)
  • Has specialized agents for different tasks (email, calendar, research)
  • Runs automated workflows on a schedule
  • Gets better over time through feedback
  • Has access to your actual tools and data (with your permission)

Think of it as the difference between a generic customer service line and a dedicated executive assistant.

What’s an “agent”?

An agent is a specialized version of Claude with a specific expertise. Instead of one AI that does everything, the system has many agents that each do one thing well:

  • An email agent that’s expert at triage and drafting
  • A calendar agent that manages scheduling
  • A research agent that gathers information
  • A writing agent that matches your voice

Each agent has its own instructions and expertise, just like team members on a well-organized team.

What’s a “workflow”?

A workflow is a sequence of steps that accomplishes a goal. For example, “morning brief” is a workflow:

  1. Check calendar for today’s events
  2. Review overnight emails
  3. Check deadlines and follow-ups
  4. Generate a summary

You can trigger workflows manually (by typing a command) or set them to run automatically.

Does it work offline?

Partially. The system needs an internet connection to use Claude’s AI for reasoning and decision-making. However, locally stored data (your vault, profile files, saved commands) are always accessible. If you lose internet, you can’t run AI-powered workflows, but your data is safe and waiting.

Troubleshooting

Claude doesn’t seem to know who I am

Your profile isn’t loading. Try:

  1. Check that profile files exist in your vault’s Self_Context folder
  2. Make sure the files have actual content (not empty templates)
  3. Try telling Claude to read your profile explicitly

A command isn’t working

Common fixes:

  1. Check that the command file exists in your commands folder
  2. Make sure the file has the right format (it should be a .md file)
  3. Try running claude and describing what you want — the system might suggest an alternative

The morning brief is empty or missing

The nightly pipeline might not be running:

  1. Check if your Mac was awake at the scheduled time (laptops in sleep mode skip scheduled tasks)
  2. Verify the automation is set up correctly (ask Claude to check)
  3. Run the morning brief manually to verify it works

Something broke after an update

Roll back to the previous version:

  1. Check your recent changes with git status
  2. If needed, restore from your last backup

For more issues, see the full Troubleshooting page.

Still Have Questions?

Open Claude Code and ask directly:

claude
I have a question about [topic]. Can you help me understand how it works?

Claude has access to the full system documentation and can explain anything in context.

Last updated on
Troubleshooting