Personal AI Assistant: How to Build One on Zo Computer

If you’ve searched “personal AI assistant”, you probably want something concrete: an AI that can do work on your behalf—not just chat.

Zo Computer is a good fit for this because your assistant runs on a real server that has:

  • your files (for memory and state)

  • a browser it can use when needed

  • integrations (email, calendar, Drive, Notion, etc.)

  • Agents (scheduled tasks) that run automatically

This tutorial walks you through building a practical personal assistant you can extend over time.

What we’re building

A “personal AI assistant” is usually three things:

  1. A reliable place to store context (files and notes)

  2. Tools it can use to take action (apps + web + code)

  3. Automations so it works while you’re not watching (Agents)

We’ll build a minimal version first, then add “memory” and automation.

Prerequisites

  • A Zo Computer workspace (where your files live and your assistant runs)

  • Optional (but recommended): connect apps in Zo settings (Gmail, Google Calendar, Google Drive, Notion, etc.)

If you don’t connect any apps, you can still build a useful assistant that:

  • reads and writes files

  • browses public webpages

  • runs scripts and services

Step 1: Pick one assistant job (don’t start broad)

Most “personal assistants” fail because they start with: “be my life copilot.”

Pick one job with a clear output. Examples:

  • “Every morning, send me a daily briefing.”

  • “When I paste a URL, save it and extract action items.”

  • “Every weekday, summarize new important emails.”

A good rule: if you can’t describe success in one sentence, it’s too broad.

Step 2: Create a dedicated folder for assistant memory

Create a folder in your workspace for the assistant’s state. For example:

  • Assistant

Inside it, keep a few simple files:

  • profile — your preferences, timezone, recurring projects

  • contacts — the people you talk to + what matters about each

  • rules — constraints (“never email anyone without asking”)

  • state.json — machine-readable state (counters, last run timestamps)

Why this works: scheduled Agents need persistent state between runs, and files are the simplest form of memory.

Step 3: Write a reusable “assistant prompt”

Create a prompt that defines how your assistant should behave. Prompts are reusable instructions you can run again later.

Your prompt should:

  • state the assistant’s job

  • list the files it should treat as memory

  • specify output format

  • define what it must never do

A good pattern:

  • Inputs: where it reads from (files, inbox, URLs)

  • Actions: what tools it can use (web, calendar, email, file edits)

  • Output: what it writes (email, file, summary)

You can start with a daily briefing prompt like:

  • Read profile and state.json

  • Check today’s calendar

  • List 3 priorities, 3 risks, and 3 suggested next actions

  • Update state.json with last-run timestamp

(You’ll connect the calendar in the next step.)

Step 4: Connect your tools (apps + web + files)

A personal assistant is only as useful as the tools it can use.

On Zo, the most common “assistant tools” are:

  • Agents (scheduled runs) ^1

  • Web browsing tools (fast extraction vs full browser) ^2

  • App integrations (Gmail, Calendar, Drive, Notion)

  • File edits (notes, plans, logs)

If your goal is “daily briefing”, connect Google Calendar.

If your goal is “inbox triage”, connect Gmail.

If your goal is “research assistant”, connect Drive/Notion for saving outputs.

Step 5: Turn it into an Agent (so it runs on its own)

Now convert your assistant job into an Agent that runs on a schedule.

Example: Daily briefing agent

  • Schedule: weekdays at 7:30am

  • Output: email (best for something you’ll search later) or SMS (best for urgent)

  • Instruction (high-level):

    • “Check today’s meetings. Create a short briefing. Include:

      • top 3 meetings + why they matter

      • 3 follow-ups I should not miss

      • 3 suggested tasks for today

      • any conflicts or travel time issues Update state.json and append a log entry to log."

The Agent page is the core feature that makes a “personal assistant” feel real: it keeps working without you remembering to ask.

Step 6: Add guardrails (so it’s safe)

Give your assistant explicit boundaries in rules.

Useful guardrails:

  • Never send an email to anyone except me.

  • Never spend money.

  • Never delete files.

  • If unsure, write a draft to a file instead of taking action.

A practical pattern is “draft-first”:

  • draft emails to <date>-email

  • create a to-do list in todos

Step 6.5: Make it private (run a local model)

If “personal AI assistant” means private AI assistant to you, the cleanest upgrade is to run a local LLM on your Zo and use it for sensitive work.

The difference is simple:

  • Hosted models: fastest to start, but your prompts leave your server.

  • Local models: lower marginal cost and more privacy, but you manage the model runtime.

Zo can host the model runtime as an always-on service, and your assistant can then treat it like another tool it can call.

If you want the step-by-step setup, follow: https://www.zo.computer/tutorials/how-to-run-a-local-llm-on-zo-computer-ollama

Step 7: Extend gradually (memory + more jobs)

Once your first job is working reliably, add a second job.

Good second jobs:

  • Weekly review: summarize what happened + what to change next week

  • Inbox follow-up checker: find threads you haven’t replied to

  • “Watch this page” monitor: check a URL daily and report changes

The key is that each job gets:

  • a clear input

  • a clear output

  • a place to store state (files)

Common pitfalls

  • Too broad: “manage my life” → start with one job.

  • No persistent memory: if you don’t write state to files, every run starts from zero.

  • No boundaries: assistants without guardrails will eventually do something you didn’t want.

Summary

A real personal AI assistant is a workflow that runs on infrastructure you control:

  • store context in files

  • connect tools (apps + web)

  • automate with Agents

Once you have one assistant job working end-to-end, you can grow it into a full “assistant system” without rewriting everything.