Personal AI Starts With Everyday Life
By Clifford · Product research
A product perspective from Memex Lab. The examples below are illustrative; future agent workflows are design possibilities, not announcements of shipped features.
Some evenings, the most useful thing an AI could do is help you finish a thought. You have a photo from a walk, a voice note from the train, and a sentence you want to keep. None of them comes with an instruction to execute.
Much of the promise of personal agents centers on delegation: research this, arrange that, keep a project moving. Those are valuable jobs. But a personal life may offer only a few of them in a month, with long stretches in between.
We are building Memex around what happens in those stretches: recording a life, returning to a conversation, and keeping the details that make occasional requests personal.
A personal life has an uneven queue of tasks
A family coordinating several calendars may have plenty to delegate. Someone else might want help comparing a purchase once a year or organizing a trip every few months. A product designed around a constant flow of assignments can feel busy in a life that has no such flow.
There is also a cost to handing something over. You explain the job, connect the right tools, check the answer, and review consequential actions. Delegation is worthwhile when that effort is smaller than doing the work yourself. Until you trust a particular workflow, keeping an eye on it belongs in the calculation.
This makes frequency a design question. An occasional task can matter a great deal, yet provide little reason to open an app on an ordinary Thursday. Personal AI needs something useful to offer on Thursday too.
Keep the afternoon before you plan the next trip
Consider an imagined weekend away. You save a picture of a quiet riverside path and a short note: “The best part was leaving the afternoon open.” A month later, another trip leaves you with a different observation: “Changing hotels every night was tiring.”
Neither note needs to become a task. You may simply want to remember the afternoon or make sense of why the second trip felt rushed. A record can keep the photograph and your words together. A conversation can help you say more about them.
These are useful outcomes already. A personal archive should be worth keeping even if you never ask an agent to arrange another holiday.
Companionship needs room for an unfinished thought
For Memex, companionship means making room to return to the things you have shared. Its documented companion features include AI characters that comment on timeline cards and chat with context from cards, knowledge, and remembered interactions. You can shape a character’s personality instead of receiving one fixed voice.
In the weekend example, a useful response might ask what made the open afternoon enjoyable. That question gives you somewhere to continue. It does not have to become a recommendation, a reminder, or a plan.
A model’s reply can also miss the point. Perhaps you enjoyed that afternoon because of the person you were with. Keeping the original words alongside the interpretation lets you disagree and explain. The conversation serves the record; it does not get the final word on what your life means.
The occasional request has a long background
Months later, you want to plan a short trip. You can pull out those two notes and explain what you liked and what wore you out. An agent given that material could ask, “Would you like to stay in one place and leave an afternoon free?” The background is already available when the occasional request arrives.
The notes also preserve the reasons behind the suggestion. You can open them, decide whether they still apply, and say, “This time I want a busy city break.” A dated observation should be evidence for a question, rather than a permanent label for a person.
This is how we think infrequent agent needs can emerge naturally from a recording habit. The request starts with something you actually want to do. Relevant records can help explain it. A quiet month of journaling still has value when no request arrives.
How personal-agent designs handle records and delegated work · Build a weekly review with selected records
Context you can keep as models change
Memex keeps primary records on your device and organizes material into cards and knowledge files. The app is open source under GPL-3.0, with configurable supported model connections, custom agents, prompts, and Skills. Those choices matter when a personal record is expected to outlast the model that first helped organize it.
Local storage and model inference are separate. A cloud provider receives the context sent in a request; optional proxy access, sync, transcription, and external tools can add other data paths. Choosing an endpoint is part of deciding where a conversation goes.
You might change models because a different one follows your writing more closely, or simply choose to reread the notes yourself. Keeping a useful archive lets those choices happen around the same material. Open source and model configuration give you ways to shape that experience; they do not guarantee compatibility with every future agent.
Supported model configuration · What local-first storage provides · Memex privacy policy
A future agent still needs your decision
The travel example describes a possible future workflow, not a claim that Memex currently books trips from your journal. Putting it into practice would require an integration, a way to select relevant records, and clear permission for what the agent may read or do.
More context is useful only when it belongs in the task. A travel request might need the two trip notes. It need not expose unrelated private conversations. Being able to choose and inspect that material matters as much as having a large archive.
For now, the product has a straightforward job: make it worthwhile to keep the photo, say the unfinished thought, and return to it. Open the riverside photograph beside the note. You still have your own account of why the afternoon mattered.