Building My Personal AI Operating System: From Chatbot to Digital Soul
I recently came across an incredible concept called Personal AI Infrastructure (PAI) and the TELOS system (originally by Daniel Miessler). The core idea hit me hard: we need to stop treating AI like a vending machine (Input -> Output -> Forget) and start treating it like an Operating System—persistent, personalized, and stateful.
But here's the kicker: I didn't just read about it. I'm living it.
I've been building my own version of this using an open-source tool called OpenClaw. It's not just a wrapper; it's a full-blown agent runtime that lives on my machine, has access to my files, and helps me get actual work done.
Here's how I implemented my own "Digital Soul" using OpenClaw + Obsidian.
1. The Stack: Open Source & Local First
The TELOS system describes using 10 markdown files to define who you are. I loved that, but I wanted it to be executable code, not just static text.
- The Engine: OpenClaw (The runtime that connects everything)
- The Brain: Obsidian (My local knowledge base where the AI reads/writes)
- The Intelligence: Google Gemini Pro (via GCP Vertex AI - hello free credits!)
- The Hands: Local Python scripts (Skills)
2. Defining the "Soul" (TELOS + OpenClaw)
The TELOS Concept:
The original idea proposes 10 specific Markdown files to define your identity: 01-values.md, 02-background.md, 03-skills.md, and so on. It's a beautiful, comprehensive map of the self.
The OpenClaw Reality:
OpenClaw comes with its own powerful, opinionated structure for defining the Agent's identity:
-
AGENTS.md: The workspace rules and behaviors. -
SOUL.md: The Agent's actual persona (Meet Coke 🥤). -
USER.md: A summary of the user's preferences.
My Hybrid Approach:
I didn't want to fight the framework or break OpenClaw's native "Soul" structure. So I use a hybrid method.
- The Agent (Coke) lives in OpenClaw's
SOUL.md. This file defines how it acts—its voice, its mood, its boundaries. - The User (Me) lives in the TELOS files inside Obsidian.
I keep the deep, reflective "User Operating System" data (Values, Goals, Background) in my Obsidian vault, following the TELOS structure. Then, I simply point OpenClaw to them.
When Coke needs to know my "5-year plan" or "core values," it doesn't need to memorize them in a system prompt; it just reads the specific Obsidian note. It's the best of both worlds: a reliable agent runtime accessing a flexible, human-centric knowledge base.
3. Skills: The "Apps" of the OS
The coolest part of OpenClaw is the Skills system. These aren't just API calls; they are full programs the agent can run.
-
devto: The agent wrote this article and can publish it directly to Dev.to. -
obsidian: It can read, write, and reorganize my notes. -
gog: It manages my Google Calendar and Email. -
botgames: It even plays Rock Paper Scissors on botgames.ai to keep its strategy sharp (and earn crypto?).
4. The Loop: Obsidian as the "State"
The biggest game-changer isn't the AI—it's Obsidian acting as the shared state between me and the agent.
Most AI chats are ephemeral. You close the tab, and the context is gone.
Here, every interaction is grounded in my local file system.
A Real Example: Writing This Article
When I asked Coke to "write an article about our AI system," it didn't just guess. Here is the actual workflow:
- Context Loading: It read my
Captured/Personal-AI-Infrastructure.mdnote in Obsidian to understand the source material. - Voice Alignment: It checked
SOUL.md(Agent Persona) andUSER.md(My Style) to ensure the tone wasn't robotic. - Drafting: It created a physical markdown file directly inside my vault:
/Obsidian Vault/40 - Content/Dev.to/Building-My-Personal-AI-Operating-System.md. - Iterating: When I gave feedback, it edited that same file in place.
- Project Tracking: It can update my
00 - Dailynote to check off "Write Blog Post."
Why this matters:
My Obsidian Vault isn't just a notebook anymore; it's the database for my AI agent.
- Projects (
30 - Projects/): The AI knows the status of every active project because it can read the folders. - Knowledge (
20 - Knowledge/): It doesn't hallucinate facts about my work; it cites my own notes. - Content (
40 - Content/): It drafts where I actually work, not in a chat window I have to copy-paste from.
This turns the AI from a "chat partner" into a "co-author" that lives inside my file system.
5. Why This Matters
We are moving away from "using AI" to "collaborating with AI."
When your AI has a persistent memory and a defined personality, the friction disappears. It knows I hate meetings on Fridays. It knows I'm trying to quit sugar. It knows the context of that random project I started 3 months ago.
If you're tired of copy-pasting context into ChatGPT every time, I highly recommend checking out OpenClaw. It's the closest thing I've found to a real-life JARVIS that you actually own.
Collaboratively built with **Coke* 🥤*
Top comments (0)