the task continues after the window closes
Google’s Agent Executor, Screenpipe recovery fixes, and ReAdapt’s relationship-state benchmark.
[!] OPERATING LAYER / PLAIN-TEXT SOVEREIGN OS
build your
personal ai os
self.md maps the layer above the model: memory, tools, connectors, local runtimes, browser agents, coding workbenches, receipts, and boundaries.
Start with the operating problem. Pick the smallest stack that can act, remember, and prove what it did.
[!] OPERATING LOOP
- capture
- route
- act
- verify
- remember
Personal AI OS
The map for tools, memory, receipts, workflows, registry entries, and stack recipes.
What is a personal OS?
The concept before tool shopping begins.
Recurring system patterns
Layers, loops, custody, boundaries, and receipts.
■ 02 / OPERATORS
systems have operators,
taste, and scars
Working patterns from people who built the tools, documented the loops, and kept humans in control.
OPERATOR / 01Context surfaces
Maggie Appleton
Digital gardens as working context, not content furniture.
OPERATOR / 02Browser memory
Linus Lee
Research loops that turn pages into reusable working context.
OPERATOR / 03Coding agents
Sergey Ris
Vibe coding as an operator workflow with taste and review.
OPERATOR / 04Evaluation
Omar Khattab
Prompts compiled into programs that can be inspected and improved.
OPERATOR / 05AI-first building
Andrej Karpathy
Software in the AI era: vibe coding for prototypes, three-layer workflow for production.
OPERATOR / 06Factored cognition
Andreas Stuhlmüller
Breaking complex AI reasoning into verifiable subtasks with human oversight.
OPERATOR / 07CLI prototyping
Simon Willison
Small tools, SQLite logs, and rapid experiments that keep the loop inspectable.
OPERATOR / 08Skill libraries
Jim Fan
Agents that learn across virtual and physical worlds through reusable skills.
■ 03 / CURRENT SIGNALS
what changed
in the field
Short, source-backed readings from the edge of personal AI systems.
Showing 1–3 of 8 signals
■ 04 / LIBRARY
choose the layer
you need
Every surface has explicit boundaries, code snippets, and evidence.
LAYER 01
LAYER 02
LAYER 03
LAYER 04
LAYER 05
LAYER 06
Concepts
Mental models, architecture, context engineering, and failure modes.
Guides
Practical step-by-step implementations with tests and diffs.
Skills
Installable agent toolpacks, Claude Code plugins, and MCP bundles.
Tools
Control surfaces, local runtimes, memory engines, and coding workbenches.
People
Operators running AI as their daily OS with public recipes.
Signals
Weekly radar dispatches on agent protocols, memory, and tooling.