AI Executive Assistant
A personal AI system that handles my email, calendar, errands, and side projects, within rules I set.
How it fits together
Solid lines carry my requests. Dashed ones mark outside input, which is screened first and treated as information.
- 8scheduled jobs
- 10+app connectors
- 2AI assistants, kept apart
- 3ways I reach it
- Voice buttonA shortcut on my phone’s home screen. Tap, speak, done.
- EmailRequests and replies.
- Claude appsOn my phone and computer, for bigger jobs like this website.
- Public emailHandled by a separate screening assistant.
- Other AI agentsWhat they send back is checked before it’s used.
- Standing rulesWhat it may do alone, and what needs me
- MemoryMy preferences and decisions, on every device
- Knowledge baseMy projects and conversations
- Activity logA record of what it did, and why
- Calendar
- Documents
- Notes
- Team chat
- Code
- Websites
- Product tools
- Travel search
- A real web browserFor forms, bookings, and checks
- My computerFiles and safe fixes, within limits I set. Anything deeper comes to me as a recommendation.
- My phoneReached from my computer
- Morning briefingWhat’s on, and what needs me
- UpkeepKeeps its knowledge base and memory up to date
- ChecksWatches my sites, and my details online
- Research teamsGroups of agents for deep research
- Independent reviewersFresh agents double-check important work
- Morning briefing
- Drafts to approve
- Calendar updates
- Alerts on my phone
- Updates to my sites and apps
- A report from every job
I wanted an assistant that does real work, not one that just answers questions, so I built one. It works across my email, calendar, documents, and notes, drives a real web browser, my computer, and my phone, runs jobs on a schedule, and remembers how I like things done.
Claude is the brain, but the system is more than Claude: it also works through my phone, my computer, and a few small tools of my own.
Connecting the apps was the easy part. The real design work was deciding what it may do on its own: it finishes tasks without checking in, and it stops only for the steps that should stay mine.
What it does for me
The jobs it handles in a normal week.
-
Email
Sorts my inbox, drafts replies for my review, and sends from its own address once I approve, with a copy to me.
-
Scheduling
Checks my calendar, finds times that work, and books appointments once I’ve said yes.
-
Morning briefing
A briefing lands in my inbox before I start the day: what’s on, the replies I owe, and anything it flagged.
-
Knowledge base
Keeps a private record of my projects and conversations up to date, so it starts every task with context.
-
Websites and apps
Built this site and my two apps, and ships changes to them.
-
Research
Looks things up on the web when I ask, and runs deep research with teams of AI agents, like the archive work behind my family history site.
How it compares to existing assistants
I didn’t build this for nothing. Existing AI assistants each cover part of the job, but none did all of this under rules I set, so I built my own.
Where it works
Usually one inbox, or one company’s apps
Email, calendar, documents, a real web browser, my computer, my phone, and jobs that run on a schedule
What it makes
Answers, drafts, and bookings
Those, plus working software: it built this website and my two apps, and ships their updates
What it knows
What you tell it, plus whatever memory the vendor keeps
A private knowledge base of my projects and people, which I own and can read
Who sets the rules
The vendor’s defaults, plus a few settings
Rules I wrote: what it does alone, what needs my yes, and what stays mine
Who can reach it
Often one broad login, and it reads mail from anyone
Split in two: the assistant strangers can email can’t open my accounts, and jobs that run on their own can only read and report
Paper trail
Whatever the app shows you
Its own email address, a copy of everything it sends, and a report from every job that runs on its own
Security built in
An assistant this connected needs firm rules. These are the ones it runs on.
-
Reading isn’t obeying
Emails, web pages, files, and messages can supply facts for a task I gave it. Only I decide what the tasks are.
-
Anything odd is flagged
If something it reads looks off, like instructions planted in a web page, it tells me where it came from.
-
It asks before speaking for me
Nothing goes to anyone else in my name, not even a draft, until I say yes.
-
Some steps stay mine
Passwords, security codes, CAPTCHAs, payment details, and permanent deletes. It opens the right page, highlights the button, and waits.
-
Scheduled jobs stick to a short list
Jobs that run on their own read and report back to me. Anything more, like preparing drafts for my approval, is on a short list I approved.
-
Other agents get need-to-know
When it works with other AI agents, it shares only what a task needs, and it checks what they send back.
-
Web searches have limits
It searches only for my own requests, keeps private details out of every search, and stops after a set number of searches and pages.
-
Web research is read-only
While researching, it never fills in forms, signs in, buys, or downloads. Anything it would do based on what it found waits for my yes.
-
Keys stay with me
I add passwords and API keys myself, and it never copies a login out of the browser.
-
Outside requests are screened
Email from people I don’t know goes to a separate assistant first, and nothing it passes on happens without my OK.
Try it
Email my assistant
Write to hello@corynitschelm.com. Public mail goes to a separate AI assistant, not the one that runs my accounts, and it has its own rules for talking to people it doesn’t know.
- It replies within a few minutesYou get a short personal note saying it’s on it.
- Public facts onlyIt answers simple questions from what’s on this site, and it never shares private details.
- Anything else comes to meMeetings, jobs, and anything personal get a personal reply from me.
- I see everythingI see every conversation.