Hallunot is a tool that helps developers pick library and framework versions that a given LLM is more likely to know well — reducing hallucinations when coding with AI without extra context (no RAG, no web search, no MCP).
LLMs hallucinate APIs, methods, and patterns — especially for library versions they weren't trained on. But the problem goes deeper than training cutoff dates alone. A model's coding ability, context limits, training openness, and a library's own stability and complexity all affect how reliably an LLM can work with a given version.
Hallunot combines library-level and model-level signals into a single heuristic score for every library + version + LLM combination: Final = LCS x LGS.
How well-suited is this version for AI-assisted coding?
- Stability — Fewer breaking changes and a mature release history signal reliability.
- Simplicity — Simpler APIs are easier for any model to get right.
- Popularity — Widely used libraries appear more often in training data.
- Language Affinity — How well the library's language/ecosystem is represented in training corpora.
- Recency Risk — Versions released near or after the model's knowledge cutoff carry more risk.
How capable is this model at coding tasks in general?
- Capability — Breadth of model features (reasoning, tool calling, structured output, multimodal I/O, etc.).
- Context Limits — Larger context windows help with complex codebases.
- Recency — More recently trained models have broader library coverage.
- Openness — Open-weight models with known training data get a bonus.
The final score is color-coded as green (low risk), yellow (medium), or red (high risk).
| Layer | Technology |
|---|---|
| Framework | Next.js 16 |
| Language | TypeScript |
| Styling | Tailwind CSS + Radix UI |
| Data Sources | Libraries.io API + models.dev |
| Testing | Vitest |
The project follows Clean Architecture and Domain-Driven Design (DDD) principles to ensure the logic is testable and maintainable:
- Domain Layer: Pure TypeScript entities and services (scoring logic). No dependencies on the database or framework.
- Infrastructure Layer: External API adapters (Libraries.io) and data mappers.
- Application Layer: Next.js Route Handlers (API) and Server/Client Components (UI).
src/
├── domain/ # Pure business logic & types
├── infrastructure/ # API adapters & data mappers
├── components/ # Reusable Radix/Tailwind UI
└── app/ # Next.js 16 pages and API routes
- Node.js 22+
-
Clone the repository:
git clone https://github.com/marcosmoraisDev/hallunot.git cd hallunot -
Install dependencies:
npm install
-
Configure Environment: Copy
.env.exampleto.envand provide yourLIBRARIES_IO_API_KEY(get one here). -
Run Development Server:
npm run dev
Hallunot provides educated heuristics, not official support indicators. The scores are designed to reduce hallucinations when using LLMs without additional context (no Context MCPs, no web search). Use these scores as a guide, not an absolute rule.
Contributions are welcome! See CONTRIBUTING.md for setup instructions, PR guidelines, and coding standards.
This project is licensed under the MIT License — see the LICENSE file for details.