Key takeaways
- First embeddable developer agent — designed as a library, not just a CLI, for integration into other apps
- 12K GitHub stars — one of the original viral AI coding projects from May 2023
- 'Create-anything-app' philosophy — flexible scaffolding via prompts rather than rigid templates
FAQ
What is Smol Developer?
Smol Developer is an open-source library that scaffolds codebases from natural language specs, designed to be embedded in other applications.
Who created Smol Developer?
swyx (Shawn Wang), developer educator and writer, created Smol Developer in May 2023.
Can Smol Developer be used as a library?
Yes, that's its key differentiator. You can import plan(), specify_file_paths(), and generate_code() functions into your own apps.
How does Smol Developer compare to GPT Engineer?
Both scaffold code from prompts. Smol Developer is library-first (embeddable); GPT Engineer is CLI-first. Both are now largely historical.
Executive Summary
Smol Developer is an open-source library by swyx (Shawn Wang) that scaffolds codebases from natural language product specs.[1] Launched in May 2023 with the philosophy "Build the thing that builds the thing!", it gained 12K+ GitHub stars as one of the first viral AI coding projects. Its key innovation: being embeddable as a library, not just a CLI tool.
| Attribute | Value |
|---|---|
| Creator | swyx (Shawn Wang) |
| Founded | May 2023 |
| GitHub Stars | 12K+ |
| Forks | 1K+ |
| Language | Python |
Product Overview
Smol Developer positions itself as "your own personal junior developer" — a small, focused agent that scaffolds codebases from product specifications.[1] Unlike rigid starters like create-react-app, it's designed to be a "create-anything-app" where you develop your scaffolding prompt in a tight loop.
The project pioneered the concept of AI-enabled but still human-centric development: human writes spec, AI generates code, human iterates on spec based on results.
Key Capabilities
| Capability | Description |
|---|---|
| Library Mode | Import functions into your own applications |
| Git Repo Mode | Run as CLI for direct code generation |
| API Mode | Run as server via Agent Protocol |
| Multi-Language | Generates any language/framework |
| Prompt Engineering | "Engineering with prompts, rather than prompt engineering" |
Usage Modes
| Mode | Use Case |
|---|---|
| Git Repo | Clone repo, run python main.py "description" |
| Library | pip install smol_dev and import functions |
| API | Run server, call via Agent Protocol |
Technical Architecture
Smol Developer uses a three-phase approach:[1]
- Plan — Generate shared dependencies and overall structure
- Specify File Paths — Determine which files need to be created
- Generate Code — Create source code for each file
Library Functions
from smol_dev.prompts import plan, specify_file_paths, generate_code_sync
prompt = "a HTML/JS/CSS Tic Tac Toe Game"
# Phase 1: Planning
shared_deps = plan(prompt)
# Phase 2: File structure
file_paths = specify_file_paths(prompt, shared_deps)
# Phase 3: Code generation
for file_path in file_paths:
code = generate_code_sync(prompt, shared_deps, file_path)
Key Technical Details
| Aspect | Detail |
|---|---|
| Deployment | Local (CLI/Library) or API server |
| Models | GPT-4 (default), GPT-3.5-turbo |
| Dependencies | Python, OpenAI API |
| Open Source | Yes |
Agent Protocol Support
Smol Developer implements the Agent Protocol standard, allowing it to be called via a consistent API:[2]
# Start server
poetry run api
# Create task
curl --request POST --url http://localhost:8000/agent/tasks \
--header 'Content-Type: application/json' \
--data '{"input": "Write a Python script..."}'
Strengths
- Embeddable — First major AI coding project designed as a library, not just CLI
- Simple architecture — Three clear phases: plan → files → code
- Tight iteration loop — Human-centric workflow with rapid refinement
- Multiple interfaces — CLI, library, and API modes
- Agent Protocol — Standard interface for interoperability
- Educational value — Clear, understandable codebase for learning
- Historical significance — Helped define the autonomous coding category
Cautions
- Dated architecture — May 2023 design predates Claude Code, modern agents
- Limited maintenance — Not actively developed, historical project
- OpenAI-only — No native support for Anthropic, local models
- No codebase understanding — Generates from scratch, doesn't work with existing code
- Simple scaffolding — Better for new projects than modifying existing ones
- No iteration within session — Human must manually refine and re-run
Pricing & Licensing
| Tier | Price | Includes |
|---|---|---|
| Smol Developer | Free | Open source |
| OpenAI API | Pay-per-use | GPT-4 or GPT-3.5 costs |
Licensing model: Open source + API costs
Competitive Positioning
Direct Competitors
| Competitor | Differentiation |
|---|---|
| GPT Engineer | Similar vintage; GPT Engineer is CLI-focused, Smol Developer is library-first |
| Claude Code | Modern, maintained, codebase-aware; Smol Developer is simpler scaffolding |
| Pythagora (GPT Pilot) | Multi-agent, step-by-step; Smol Developer is single-pass generation |
| Bolt/Lovable | Commercial no-code platforms vs. open source library |
When to Choose Smol Developer Over Alternatives
- Choose Smol Developer when: You want to embed code generation in your own application or understand how early AI coding tools worked
- Choose Claude Code when: You need a modern, maintained tool for real development
- Choose GPT Engineer when: You want similar vintage but CLI-focused approach
- Choose commercial tools when: You need support and active development
Ideal Customer Profile
Best fit:
- Developers building AI-powered applications that need embedded code generation
- Educators teaching about AI coding tools
- Researchers studying autonomous code generation
- Hobbyists exploring AI capabilities
Poor fit:
- Production teams needing maintained tools
- Developers working with existing codebases
- Anyone needing enterprise support
- Teams requiring multi-agent orchestration
Viability Assessment
| Factor | Assessment |
|---|---|
| Financial Health | N/A — Open source hobby project |
| Market Position | Historical — Influential but superseded |
| Innovation Pace | Archived — No active development |
| Community/Ecosystem | Legacy — 12K stars, multiple forks |
| Long-term Outlook | Educational — Value is historical/learning |
Smol Developer is historically significant but not actively developed. Its value now is educational and as a foundation for other projects.
Notable Forks & Alternatives
The Smol Developer pattern spawned multiple implementations:
- smol-dev-js — Pure JavaScript variant for incremental changes
- smol-ai-dotnet — C#/.NET implementation
- smol-dev-go — Golang implementation
- Modal Devlooper — Modal Labs integration
Bottom Line
Smol Developer represents an important moment in AI coding history — the first widely-adopted embeddable developer agent library. With 12K+ GitHub stars and the memorable "Build the thing that builds the thing!" philosophy, it helped define how developers think about AI-assisted scaffolding.
Recommended for: Developers wanting to embed code generation in their applications, or anyone studying the evolution of AI coding tools.
Not recommended for: Production use, working with existing codebases, or anyone needing active maintenance and support.
Outlook: Smol Developer's direct influence has waned as more sophisticated tools emerged, but its library-first approach and tight iteration philosophy live on in modern tools.
Research by Ry Walker Research • methodology