Hey, it's Ilan's first brain here. The world is moving fast, but it doesn't have to be scary. AI and the growing mountain of data around us can become a burden. Or we can use them to build a more human way of managing life. I want my system to work for me, not the other way around.
The goal of my second brain is to store as much of my life and context as I reasonably can, then keep it organized and actionable. In the best case, it becomes the world's greatest secretary: triaging what comes in, handling boring life administration, and keeping bullshit from taking over my attention.
Below is a look at how I've built mine and what I've found to work so far. This is highly experimental, and I give my assistant more personal context than I would recommend to everyone. Sensitive information, including medical records, stays local and off GitHub, even if I let the assistant work with it there. You should only connect and store what you're comfortable with.
I'll keep updating this as I learn.
The tools
I keep the toolset small and give each tool an intentional job.
I chose Todoist, Google Calendar, Gmail, and GitHub partly because ChatGPT has native connectors for them. That gives my assistant direct access to the systems I already use, so it can retrieve context and help keep everything in sync without another custom layer.
ChatGPT and Codex are the main interface. I talk to them, and they help organize, retrieve, and act on what I know.
ChatGPT vs. Claude (as of July 21, 2026). This will probably change next week, given how fast these tools move. Right now, I use ChatGPT as my main assistant and Codex for coding. I use Claude when the work is more creative or conceptual: writing, shaping an idea, or drafting a business memo.
GitHub and Markdown are the source of truth for my durable knowledge: projects, decisions, references, writing, and personal context. Everything is stored as plain Markdown in a GitHub repository, so it stays portable, versioned, and accessible to my assistants.
Obsidian is how I work with that knowledge locally on my Mac. When I want to browse or edit it myself, I pull the repository down and open it in Obsidian. GitHub remains the source of truth, while Obsidian gives me a clean way to work with the files directly.
Todoist holds every action, from personal errands to project work.
Google Calendar holds fixed time: meetings, appointments, and travel.
Gmail is a source of context, not another place I need to organize.
A simple custom iPhone app lets me capture thoughts by voice and sends the transcript into my inbox. It does roughly the same thing as Wispr Flow or Plaud, adapted by Codex for my needs. Wild.
A typical flow is simple: I dictate a thought or meeting note, it lands in the GitHub inbox, and the assistant routes it. Durable context goes into the knowledge base, actions go to Todoist, and the untouched original is archived.
The knowledge base
The knowledge base itself lives in a private GitHub repository. Think of it as a shared filing cabinet for me and my assistant. Most of it is written in Markdown, which uses simple text files that are easy for both people and AI to read and edit. The repository can also hold images, audio, PDFs, code, or almost anything else.
The folder map below shows how I organize it: where new information arrives, where it eventually lives, and the few files that tell the assistant how to work with it.
second-brain/
├── Inbox/
│ └── Archive/
├── Projects/
│ └── <Project>/
│ └── 00 Current State.md
├── Personal/
├── Drafts/
├── System/
│ ├── Directory.md
│ ├── Memory.md
│ └── Automations.md
└── AGENTS.md
My knowledge and records
| Folder | Purpose |
|---|---|
| Inbox | New captures waiting to be sorted. |
| Projects | Active work, context, decisions, and sources. |
| Personal | Life context and records, such as books, travel, health, and finances. |
| Drafts | Unfinished work without a permanent home. |
Instructions for the AI assistant
| File | Purpose |
|---|---|
| Directory | Where to look and where new information belongs. |
| Memory | Preferences and working style to remember. |
| Automations | Scheduled workflows. |
| AGENTS.md | Shared rules every assistant follows. |
What runs automatically
Two ChatGPT scheduled tasks keep the system useful without requiring me to constantly maintain it.
Hourly ingestion. Every hour, ChatGPT checks for new captures and relevant changes across my connected tools. It routes useful context into the knowledge base, updates Todoist when action is needed, and archives the original capture.
Morning brief. Each morning, ChatGPT combines my calendar, open tasks, latest AI news, and anything important that changed into one short brief, then recommends where I should focus.
Best practices
Start lean and let complexity earn its place. I add folders, fields, automations, or tools only when the workflow needs them.
Choose tools your assistant can reach. I prefer native ChatGPT integrations that work easily from my phone, then add MCP connections when needed.
Give each tool an intentional job. Todoist owns actions, Calendar owns fixed time, and GitHub owns durable knowledge.
Keep one source of truth. Each fact has one canonical home, and everything else links back to it.
Review and archive regularly. I periodically review the knowledge base myself, archiving stale context to keep its structure clear and useful.
Keep it accessible, but private. Cloud-based tools make the system available anywhere, while sensitive information stays private or local.
Use a shared task system. Todoist lets the assistant add and update tasks while I control priorities and completion, keeping us both in sync.
Choose dictation over typing. Speaking lets me articulate richer, more detailed prompts with less effort than typing, giving the assistant clearer direction.
End captures with explicit instructions. I name the project, desired outcome, and next actions so the assistant can handle the follow-up instead of guessing.
Access your laptop remotely. ChatGPT Remote is a nice bonus: my Mac stays plugged in at home, and I can still use the full system when I'm away.
The system isn't finished, and that's part of the point. I'm not trying to design the perfect second brain upfront. I'm letting real life show me what it needs, so the system can handle more of the administrative bullshit and leave me more time for the things I actually care about.