Project notes
An experiment for reading code together with AI
Decoder is a personal study project, open source and free. These notes describe how it is designed, why it exists and what it does not do.
Why this experiment
Generative AI can produce code that looks correct but still contains bugs or sub-optimal choices. This project is a personal exercise to study whether a second reading layer — assisted by another model and using public prompts — can help understanding it. It does not replace professional tools for static analysis, security review or human code review.
Project choices
- A study tool
Decoder is a study tool, not a certified verification system. It helps reading code but does not replace your judgement or a professional review.
- Open source (MIT)
MIT License: free to use, modify and distribute. No hidden components.
- Technical measures in place
User API keys are AES-256-GCM encrypted and are not returned to the browser. Uploaded files are isolated per user via Row Level Security. No ad trackers, no behavioural profiling.
- BYOK
You can configure your own key for the AI provider you prefer and pay it directly. Markdown export of explanations is available.
- Optional local inference
When Ollama or LM Studio is selected, AI inference runs on your localhost and the file body is not sent to external providers. Uploaded files remain in the server's storage, accessible only to you.
- Multilingual
Languages available today are English, Italian and Simplified Chinese. Further languages depend on community contributions.
- Depth levels
You can pick the depth of the explanation, from concise to technical.
Project limits
- Uploaded code is not used to train models: the author has neither the access nor the infrastructure to do so.
- No data sale, no marketing integrations: there is no commercial activity behind the project.
- No AI-provider lock-in is in place; the project is BYOK.
- Markdown export of explanations exists but remains subject to demo availability.
- Uploaded files may be deleted automatically after inactivity; long-term retention is not guaranteed.
- AI output must always be verified by a competent person: it is not professional advice nor a correctness guarantee.
Contributions
Contributions and reports are welcome via GitHub Issues and Pull Requests. The author reviews them in spare time, with no response-time commitments.
- Extend the analysis rules (src/lib/analysis-prompt.ts) to cover new smells or languages.
- Add a language by creating a folder in src/i18n/locales/.
- Integrate new AI providers or local engines compatible with the OpenAI API.
- Improve the user documentation in /docs.
Future ideas
Areas under exploration are documented in the README and in the internal roadmap file in the repository, with no deadlines and no delivery guarantees.