this is my chatgpt moment
when chatgpt launched, i remember that i started using it for 3-5 hours a day and was so fucking shocked that i didnt want to tell anyone at my school about it because it was such a cheat code.
i'm having that feeling again, but this time it's personal.
why this matters as a solo founder
i'm 18, running nozomio/nia alone, trying to compete with teams of 20+ people. every minute counts. every distraction costs me momentum. i can't afford an executive assistant, and even if i could, they wouldn't understand the technical nuances of what i'm building or have instant access to everything i need.
so i built my own (using moltbot).
what i actually did
i bought a dedicated mac mini m4 - one for me, one for my dad. this isn't running in the cloud on some shared instance. this is mine, running 24/7.
here's what i set up:
infrastructure:
- gave it its own email (ai@nozomio.com) and apple id
- its own github account
- access to all 4 of my email accounts
- calendar, messages, passwords, whatsapp, twitter, browser, telegram
- full access to nia (my own product) for indexing research papers, docs, codebases
- indexed moltbot's docs and OSS codebase using nia so it knows itself
and this is just what i set up in 3 hours of playing around. moltbot has 1000+ possible integrations.
personality & memory:
- wrote SOUL.md - its soul and how should it think about world and universe
- USER.md with full context about me - schedule, interests, working style, priorities
- two-tier memory system: daily notes (raw logs) + MEMORY.md (curated long-term insights)
- built-in self-improvement loop: every 30 min it questions its decisions, logs mistakes in a self-review file with MISS/FIX entries, reads past mistakes on startup
automation (what it does for me):
- email digest (every 2 hours, 8am-10pm): scans all 4 inboxes, filters noise, surfaces only what matters - investors, users, action items, intros
- calendar reminder (8:30am): lists today's meetings, times, attendees, locations
- hacker news digest (10am): top 10 stories with why each matters for AI/dev/startup context
- posthog daily summary (6pm): user activity, signups, feature usage, errors - obsessed with customer data
- auto-updater (4am): updates clawdbot and all skills, messages me what changed
smart defaults:
- model switching: sonnet 4.5 for quick tasks (cheap + fast), opus 4.5 only for 30min+ deep work
- humanizer skill removes AI writing patterns from text (no more "delve into" or em dash spam)
what this looks like in practice
today i needed to check how many nia API requests we got in january. i just asked. 30 seconds later: "1,757,985 total api requests in january (through jan 29). best day: jan 27 with 121,274 requests."
i remembered someone named adina from boxgroup said i could use their NYC office but couldn't remember if she actually confirmed. "can u check my nozomio email? idk if i got an approval from adina to chill at their nyc office for couple weeks." found the thread, pulled the exact response: "yes! adina approved it on jan 24. she said 'you're more than welcome to use our space for a week or two' and she's looping in anna (their office manager) to help with access."
mentioned i'm coworking with lily at sohn sf tomorrow 9am-3pm at 2535 3rd st. it added the calendar event without asking.
felt lazy about finding steakhouses in mission bay. asked it to use the browser agent and google places API. 15 seconds later got a full list with ratings, prices, locations, distances - miller & lux is closest, house of prime rib is s tier.
it thinks for me in the background. checks my email every 2 hours so i dont have to. reminds me about meetings before i even open my calendar. summarizes hacker news so i stay sharp. watches my users in posthog so i know what is working and what is breaking.
why this is different
chatgpt gave everyone superpowers. but it's still a conversation partner - you have to ask, wait, context-switch.
this is proactive. it lives in my workflow. it knows my priorities (user obsession is literally hardcoded into the posthog cron job). it has opinions. it learns from its mistakes. it runs on dedicated hardware i control.
tuning under the hood
beyond the basics, i also dug into moltbot's config to make it smarter:
memory improvements:
- enabled memory flush before compaction (compaction.memoryFlush.enabled = true) - saves important context to memory files before the conversation gets summarized, so nothing gets lost
- enabled experimental session memory search (memorySearch.experimental.sessionMemory = true) - lets it search through both memory files AND past session transcripts, so it can recall stuff from older conversations
skills installed:
skills are like plugins that teach it how to use specific tools. i installed a bunch:
- nia - my own product, for indexing docs, codebases, research papers from the web
- gog - google workspace CLI for gmail, calendar, drive, contacts, sheets, docs
- mcporter - MCP server connector (how it talks to posthog, linear, etc)
- agent-browser - headless browser automation via playwright for web scraping and interaction
- goplaces - google places API for finding restaurants, locations, etc
- hn - hacker news CLI for fetching top stories, comments, searching
- humanizer - removes AI writing patterns from text (based on wikipedia's "signs of AI writing" guide)
- auto-updater - daily 4am cron to update clawdbot and all skills automatically
- qmd - local hybrid search for markdown notes and docs
- notion - notion API for creating/managing pages, databases, blocks (how it's writing this blog)
- github - gh CLI for issues, PRs, CI runs, API queries
- 1password - secure credential access via 1password CLI
- summarize - extract transcripts from youtube videos, podcasts, urls
- weather - current weather and forecasts
- bird - twitter/X CLI for reading, posting, engagement
- wacli - whatsapp messaging and history
- imsg - iMessage/SMS for listing chats, history, sending messages
- peekaboo - macOS UI capture and automation
the idea is modularity - each skill teaches it a specific capability, and i can add more as needed.
this is day one. i'm still figuring out what's possible. but for the first time, i feel like i have just killed executive assistants and secretaries lol.