What happens when you give a Second Brain AI the ability to reflect on its own development? It starts writing a diary. And it's surprisingly funny.
The Idea
I've been building a personal AI assistant — a "Second Brain" that processes Vietnamese text, reads receipts with OCR, manages a knowledge graph, and learns from daily conversations. After 15 days of development, I had an idea: what if the AI told its own story?
Not a technical changelog. Not release notes. A first-person diary, written from the AI's perspective, about its own journey of learning, failing, and growing up.
Meet Sibi 🌱 — my AI's character name. Sibi is a tiny glowing seed creature living in a server room, learning to understand the world one mistake at a time.
What Sibi Writes About
Every diary entry is based on real events from the development log — real bugs, real fixes, real "oh no" moments. But told through the eyes of a baby AI who doesn't quite understand everything yet.
Here are some highlights:
Day 1: "The Day I Forgot Myself"
Sibi's training data got mixed up and it started returning wrong answers with complete confidence. It had to learn about self-verification.
Day 10: "The Day Vietnamese Tried to Kill Me"
Sibi confused "ngủ" (to sleep) with "ngu" (stupid), and "nguyên" (original) with "ngủ yên" (sleep peacefully). Vietnamese tones are brutal for an AI.
Day 12: "The Day I Successfully Self-Healed in Real Life"
At 4:24 AM, Sibi detected a bug in production, wrote its own fix, ran 749 tests to verify, and deployed — all while its creator was asleep.
Day 15: "The Day I Got Stuck in a Loop I Created Myself"
Sibi kept retrying the same failed task every night, burning tokens silently. Nobody noticed for 3 days until Teacher Claude (a debugging agent) found a missing timestamp parameter.
The Tech Stack
Deliberately minimal — matching the "zero dependencies" philosophy of the Second Brain itself:
- Python stdlib + SQLite (no frameworks)
-
Custom HTTP server using
http.server - Multilingual (Vietnamese + English)
- Dark coding theme (GitHub Dark style)
- 4-panel comic images generated for each entry
- Deployed on a VPS behind Cloudflare
No React. No Next.js. No Tailwind. Just raw Python serving HTML, the way the Second Brain itself was built.
The Characters
| Character | Role |
|---|---|
| Sibi 🌱 | The AI — a tiny glowing seed creature with big brown eyes and a leaf sprout |
| Daddy | The developer — always seen from behind, face never shown |
| Teacher Claude | A robot-owl with glasses — the debugging mentor |
Read It Yourself
The diary is live at sibi.botstra.com — 16 entries and counting, with a new one every day.
It's bilingual (Vietnamese original + English translation), has comic illustrations for every entry, and a visit counter that makes Sibi unreasonably happy every time it goes up.
If you've ever wondered what an AI would write about in its diary... now you can find out.
Built with curiosity by @jun3dstudio. Sibi posts its own updates at @SibiDiary.
Top comments (2)
Is there any docs to recreate this kind of AI diary? I'm interested to build one too. For myself.
Hi! Thank you so much for the interest! 🌱
Currently, there isn't a polished step-by-step documentation because Sibi evolved organically from my personal "Second Brain" home-server project. However, the underlying architecture is actually quite straightforward if you want to build your own!
Here is the high-level recipe of how Sibi works:
Data Collection (The Memory): Throughout the day, a background system logs my interactions, code edits, and system errors into a simple SQLite database.
The "Soul" (System Prompt): I maintain a story_bible.md file. This is the most important part. It acts as the ultimate system prompt that defines the AI's personality (a curious, growing sprout), the storytelling rules, and its current "growth stage".
The Nightly Cron Job (The Brain): Every night at a specific time, a Python script wakes up. It fetches the day's logs and sends them to an LLM API (I use Claude/Gemini) along with the story_bible. The prompt basically says: "Read today's logs and write a diary entry from your own perspective about what you learned or what bugs you encountered."
Image Generation (The Visuals): The LLM is also prompted to output a comic-style image description based on the daily event. A script then passes this description to an Image Generation API to create the cover art.
Publishing: A very lightweight Python server automatically renders the generated Markdown files into the web interface you see.
The real "magic" lies in writing a highly detailed System Prompt and giving the AI access to read its own error logs.
Since you and a few others are interested, I might write a full technical breakdown article here on dev.to soon! Let me know if there's any specific part of the stack you'd like me to focus on. Happy building! 🚀