← All Projects

Family Assistant

A private daily family digest combining calendar, email and captured plans.

With two children, our family seemed to have an endless stream of appointments, school activities, clubs and things we needed to remember. The information was scattered across emails, calendars and school apps, often in Spanish. We got to the point where we were forgetting things and missing appointments. Much of the responsibility for remembering and organising everything fell on one parent, my wife, and I wanted us to share that more evenly.

My initial idea was quite simple, to have an assistant read the relevant information and send us a useful email every morning, so we both knew what was coming up and what needed attention. I also wanted to help my son take ownership of his own diary, schoolwork and activities, rather than relying on us to manage them for him. That became Family Assistant, a private household tool built around daily emails and a Telegram assistant called Harriet.

Product screens and details

A couple talking over breakfast in a bright kitchen

The story behind the project

The product

Family Assistant brings together information from our personal Gmail accounts, Google Calendars and other family sources, including school information and local news. Each morning it sends us a digest of upcoming appointments, commitments and anything that needs a decision or a check. It includes school lunch menus, the weather and practical suggestions for the day, and it distinguishes confirmed arrangements from things that still need sorting out.

My son receives his own daily digest of schoolwork, tests, clubs and other activities, drawing on a calendar he can maintain himself. It also includes a link to a maths challenge and a weekly music recommendation. The idea is to give him greater ownership of his diary and commitments, rather than outsourcing the responsibility for keeping track of them to us.

A Telegram assistant provides a way to ask questions, add or correct information and manage household reminders. One practical use has been reminders for medication that has to be given several times a day. Alongside the daily emails, the system compiles local news and events from Spanish sources, translating information that I might otherwise miss or leave my wife to find. There's also a monthly email that identifies subscriptions and unsubscribe links in my inbox.

There isn't a traditional application interface. Python programmes, a SQLite database and scheduled jobs run on an AWS Lightsail server, with OpenClaw supporting the AI interactions. The system reads Gmail and calendar information, processes new messages into household records and produces emails from that information. Dates, reminders, recipients and delivery rules are checked in code rather than left entirely to a language model. The data and sending components are separated, with encrypted backups and tests around the behaviour that matters most.

Building it with AI

I started with the idea of a daily reminder email, but quickly realised that producing something trustworthy from several inboxes and calendars was more complicated than asking an AI model to summarise messages. The same appointment might appear in an email and a calendar, a booking might be pending and then confirmed, or an old message might contradict an update. I needed the assistant to recognise those relationships rather than list everything it had found.

I built the service in Python using SQLite for persistent state and small scheduled processes rather than one unrestricted AI agent. Gmail and Google Calendar provide the main information, while OpenClaw powers the Telegram assistant and parts of the message interpretation. I kept AI responsible for interpreting messages and improving the wording, but used defined application rules for dates, status changes, recipients and reminders. That separation is particularly important when the input is an email from outside the household: an email should provide information, not gain the authority to instruct the system.

The product has developed through daily use. I added a child-friendly daily digest, school lunch menus, medication reminders, local events and news, and a monthly subscription review because they addressed things we were actually dealing with at home. The same real-world use has exposed problems. For example, school menu PDFs can have inconsistent layouts, appointments can appear twice under Spanish and English names, and a confirmed booking can still be reported as outstanding when the system fails to link the two messages. I've been correcting those behaviours as they appear rather than treating the initial build as finished.

I also found that even when every individual item was technically correct, the morning email could become long and repetitive. I introduced a separate AI composition step to combine related updates and improve the writing, with rules to prevent it from dropping important dates, amounts or other protected information. Reliability, recovery and costs have all required attention: the repository records a substantial automated Python test suite, regular backups, a controlled update process and experiments with less expensive AI models.

Results and reflections

We are using Family Assistant, and the daily email has already made a practical difference, as it helps me take a more active role in family planning rather than leaving one person to remember everything. I also can't yet say we've stopped missing appointments altogether.

It is still very much a work in progress. Some information is missed, some is repeated and the assistant doesn't always understand the difference between an update, a confirmation and something that needs a decision. The next step I have in mind goes beyond reminders: recognising what needs to happen around an event, such as making sure a birthday cake or present has actually been organised, rather than simply mentioning that a birthday is coming up.

The biggest lesson so far is that a useful family assistant isn't just a chatbot connected to email. It needs a reliable understanding of who is doing what, what has already been arranged and what still needs attention. It also needs to make life easier without becoming another thing the family has to manage. Building it for our own use gives me a constant feedback loop, and I expect to keep refining it as we discover what is genuinely useful.

Have an idea or a process you want working?

I take ideas from spec to a working product, and the same approach applies to automation and AI projects inside organisations.

Contact