From prompt to codebase, with an eject button
It embeds a junior AI developer in your app to scaffold code from plain English, keeping a human in the driver's seat.

What it does
smol developer is a Python library that treats an LLM as a junior developer you can embed in your own application. Feed it a plain-English product spec and it returns a plan, a file list, and generated source code. It is designed to be interrupted: you read the output, paste errors back into the prompt, and iterate until you decide to take over the codebase yourself.
The interesting bit
Instead of monolithic generation, the pipeline splits work into three discrete stages—plan, specify_file_paths, and generate_code_sync—using OpenAI’s function-calling API to force valid JSON outputs. A middle “shared dependencies” step lets the model talk to itself about cross-file contracts before writing code, an admission that whole-program coherence is still hard.
Key highlights
- Embeddable library (
smol_dev), not just a CLI: import the planning and generation functions directly into your own app. - Human-in-the-loop by design; the README explicitly frames the workflow as “engineering with prompts, rather than prompt engineering,” and encourages you to take over when the AI stops adding value.
- Includes a
debugger.pythat reads the entire existing codebase plus your error message to suggest targeted fixes. - Also exposes an Agent Protocol-compatible API for remote task execution.
- The author is admirably candid about limitations, noting that
shared_dependencies.mdis sometimes incomplete and that the repo contains Chrome-extension-specific cruft.
Caveats
- The codebase carries some use-case-specific hacks for Chrome extension image handling that leaked into general logic.
- Cross-file dependency planning is admitted to be imperfect; the author resorted to hard-coding specific names in prompts to keep outputs coherent.
- The example gallery is still sparse; the author apologizes for being unprepared for the project’s popularity.
Verdict
Worth a look if you want to prototype scaffolding or embed spec-to-code generation into another tool, but expect to stay hands-on—it is explicitly a junior developer, not a drop-in replacement for one.
Frequently asked
- What is smol-ai/developer?
- It embeds a junior AI developer in your app to scaffold code from plain English, keeping a human in the driver's seat.
- Is developer open source?
- Yes — smol-ai/developer is open source, released under the MIT license.
- What language is developer written in?
- smol-ai/developer is primarily written in Python.
- How popular is developer?
- smol-ai/developer has 12.2k stars on GitHub.
- Where can I find developer?
- smol-ai/developer is on GitHub at https://github.com/smol-ai/developer.