Personal Software
Personal Software refers to custom, hyper-tailored software applications built by individuals or small teams to solve specific, idiosyncratic problems in their immediate personal, domestic, or operational environments. Unlike commercial, mass-market software—which requires broad generalization, generic user models, multi-tenant databases, public authentication, and complex distribution channels—personal software is optimized for an audience of one or a single household.
The paradigm emerged as AI coding agents (such as claude-code and codex) and rapid development platforms (Lovable, Replit, Bolt) collapsed the technical barriers to software construction, shifting the bottleneck from syntax mastery to clear problem specification.
The Five Software Shapes Taxonomy
As defined by nate-b-jones in nobody-laid-out-the-five-kinds-of-software-you-can-make, personal software projects map into five distinct technical shapes:
| Shape | Execution Target | Data & State Layer | Best Use Cases | Overhead / Complexity |
|---|---|---|---|---|
| 1. Local Tool | Single local machine / CLI | Local files, SQLite | Private research, document renaming, data parsing | Lowest; zero server/auth management |
| 2. Web App | Modern web browser / PWA | Supabase, Postgres, Lovable Cloud | Household task tracking, appliance logs, private dashboards | Low; recommended default for personal software |
| 3. Native Phone App | Mobile OS (iOS / Android via Expo) | Managed backend (Supabase, Firebase) | Needs deep hardware APIs (Bluetooth, NFC, background GPS) | Medium-High; involves build signing & store reviews |
| 4. Background Service | Headless server / cron / webhook | Database, message queue, email alerts | Scheduled public record scraping, webhook alerts | Low-Medium; runs silently without UI |
| 5. Hardware Edge Project | Single-board computer (Raspberry Pi, ESP32) | SQLite, Home Assistant, ESPHome | Radio signal decoding (AIS ferry tracking), sensor arrays | Medium; bridges physical and digital perimeters |
Architectural & Project Steering Principles
The 4-File Project Steering Framework
To prevent non-deterministic agent hallucinations or unwanted infrastructure sprawl, personal software builders maintain four durable markdown files at the project root:
project.md: Plain English scope, user personas, current vs. desired state, deployment target, and privacy boundaries.decisions.md: Append-only record of technical choices (database engine, auth mechanism, hosting), documenting trade-offs and rationale.scenarios.md: Concrete end-to-end user workflows and edge cases serving as automated or manual test criteria.claude.md/agents.md: Operational harness instructions enforcing plain-English explanations, credential protection, and pre-execution confirmations.
Stack Decoupling
Builders maintain long-term flexibility and data ownership by separating three layers:
- The Model: Underlying intelligence engine (Claude 3.7, GLM 5.3, GPT-5).
- The Harness: The coding CLI / agent interface (claude-code, codex).
- The Host / Storage: Durable database (Postgres/Supabase) and hosting provider (Vercel, Render, local Raspberry Pi).
Strategic Significance
Personal software represents a structural shift in software economics. For decades, technical builders lived far from real-world domestic and niche operational problems, while domain owners lacked coding ability. Coding agents bridge this gap, enabling individuals to become continuous software maintainers of their own lives and domestic infrastructure.