Reverse engineer a complex codebase into business workflows, connected code, edge cases, and evidence-backed risks.
⭐ Hit Star to help increase UnvibeCode's visibility among developers.
Unvibe complex code. Trace what the business actually does.
Reading files one by one does not show the complete business workflow. Pasting a large repository into an LLM can also lose the connections between entry points, state changes, dependencies, edge cases, and business outcomes.
UnvibeCode reverse-engineers a complex codebase into business workflows, connected code, edge cases, and evidence-backed risks.
Requires Python 3.11 or newer.
py --version
py -m pip install --upgrade unvibecode
py -m unvibecode --help
py -m unvibecode review --repository "D:\path\to\repository"Example:
py -m unvibecode review --repository "D:\Projects\customer-support-agent"python3 --version
python3 -m pip install --upgrade unvibecode
python3 -m unvibecode --help
python3 -m unvibecode review --repository "/path/to/repository"Example:
python3 -m unvibecode review --repository "/Users/yourname/Projects/customer-support-agent"python3 --version
python3 -m pip install --upgrade unvibecode
python3 -m unvibecode --help
python3 -m unvibecode review --repository "/path/to/repository"Example:
python3 -m unvibecode review --repository "/home/yourname/projects/customer-support-agent"No activation key. No customer OpenAI API key. The repository path is the only required input.
Open 00_unvibecode_results.html and explore the outputs in this order:
Start here to understand end-to-end business operations reconstructed from the code.
Trace imports, symbols, and static code relationships. Hover a file to preview directly connected code; click a file to choose how much connected code to download for use with an LLM.
Choose Narrow (~30K tokens), Optimal (~45K tokens, recommended), or Wider (~75K tokens). Larger packages include more surrounding code.
Review evidence-backed risks tied to analyzed workflows and supporting code.
Download the complete normalized repository ZIP for LLM-assisted analysis, API workflows, or a reusable analysis bundle. This is the full repository context; use Connected Code when you want a smaller selection around a file.
Tools such as Probe, Graphify, PR-Agent / Qodo Merge, and Repomix solve useful parts of code understanding. UnvibeCode goes further by making the business workflow implemented across the repository the main unit of analysis.
| Tool | Strong at | Where UnvibeCode goes further |
|---|---|---|
| Probe | AST-aware code search, extraction, and code context for AI agents | Search and retrieval help locate code; UnvibeCode reconstructs the end-to-end business workflow that crosses those files and functions |
| Graphify | Building and querying a knowledge graph of code, documents, and relationships | A graph explains how things connect; UnvibeCode additionally reconstructs business workflows, edge cases, and evidence-backed business risks |
| PR-Agent / Qodo Merge | Reviewing pull requests, describing changes, and suggesting improvements around a diff | PR review starts from changed code; UnvibeCode reverse-engineers the existing repository and its business workflows beyond a single change set |
| Repomix | Packaging a repository into AI-friendly context for LLMs | Repository context gives an LLM source material; UnvibeCode additionally reconstructs workflow logic, state changes, edge cases, and business consequences |
| UnvibeCode | Connected code + business workflows + business logic + edge cases + evidence-backed business risks | The codebase is reviewed through the business workflows it implements, not only files, graphs, diffs, or context packages |
The core unit in UnvibeCode is not a file or a diff. It is the business workflow implemented across the codebase.
Developer trials across public and personal repositories repeatedly highlighted three useful parts of the product:
- Business risks that could be independently checked: in a trial on the Rich Python library, a developer independently reproduced a Business Risk Finding surfaced by UnvibeCode.
- Business workflows instead of only repository structure: in a Django project, a developer found the Business Workflow Map useful for understanding workflows covering student records, API operations, registration, and authentication/dashboard delivery.
- Useful output even when a repository is too large for the deeper review: in a trial on a repository with about 12.6M estimated source tokens across 3,914 source files, UnvibeCode still produced the Connected Code Map and downloadable repository context while safely skipping the deeper workflow review.
UnvibeCode reverse-engineers a complex codebase into business workflows, connected code, edge cases, and evidence-backed risks.
Start with business workflows, not individual files. UnvibeCode traces each workflow to its connected code, entry points, dependencies, state changes, and edge cases so developers can understand how the system actually works.
UnvibeCode reconstructs business workflows and business logic across files and functions and connects each workflow to its entry points, decisions, dependencies, state changes, and supporting code evidence.
UnvibeCode 0.3.3 supports connected-code mapping and LLM-context preparation for:
| Language | Recognized file types |
|---|---|
| Python | .py |
| JavaScript | .js, .jsx, .mjs, .cjs |
| TypeScript | .ts, .tsx |
| Rust | .rs |
| PHP | .php |
| Ruby | .rb |
| Web assets | .html, .htm, .css |
C, C++, Java, Go, C#, Kotlin, and Swift files are detected but are not yet included in full connected-code analysis.
UnvibeCode also includes a research-backed RAG reliability guide for teams working with mixed PDFs, changing policies, live structured data, multi-hop relationships, and repository code.
- Production RAG Reliability Guide — parser routing, deterministic factual realization, knowledge graphs, version metadata, DB/Python routing, structured code retrieval, and RAG evaluation.
- Evidence-Bound Factual Repair in Retrieval-Augmented LLM Answers — related controlled preprint separating semantic evidence localization from deterministic factual realization. The paper is a preprint and is not yet peer reviewed.
- UnvibeCode RAG Review skill — reusable review instructions, checklist, and runnable synthetic regression example.
The guide distinguishes experimental findings from related literature and from engineering recommendations so implementation advice is not presented as stronger evidence than the sources support.
- Production RAG Reliability Guide — research-backed parsing, deterministic factual realization, knowledge graphs, versioning, code retrieval, and evaluation
- RAG review skill, checklist, and Python example
- Related research preprint — Evidence-Bound Factual Repair in Retrieval-Augmented LLM Answers
- Quick start for Windows, macOS, and Linux
- Understanding the four outputs
- How UnvibeCode works
- Repository limits and responsible-use boundaries
- Data processing and privacy
- Troubleshooting and support
- Public preview and public-repository reviews
- UnvibeCode Engineering Challenge 2026
- Contributor and challenge credentials
For reproducible package problems or feature requests, open a GitHub issue.
For product questions, public-repository review requests, or collaboration enquiries, email divya.singaravelu@iiml.org.
- Read the contribution guidelines before proposing a change.
- Explore the pre-built business workflow packs or contribute a new one using the MECE rules.
- Look for a focused starting point in good first issues.
- Reuse and distribution are governed by the repository's license.