UltraGXIntroduction
UltraGX reads a GeneXus Knowledge Base (KB), from version 7.5 to 18, and puts everything it learns about it into open files: Markdown documents to read, JSON to process and a browsable website that works offline. The result serves three purposes:
- Understanding the system without opening the GeneXus IDE: what each object does, which business rules it applies, which tables it reads and writes, who calls it and what it calls.
- Preserving the knowledge in a format you can version, search and share, even when the people who built the system are gone.
- Migrating: the output is designed so that an AI agent (Claude Code, Codex, Cursor, Antigravity) can regenerate the application in another technology, module by module, without interpreting GeneXus from scratch again.
What it does, step by step
- Reads the KB from an XPZ/XML export, from a text export or, on Windows with GeneXus installed, by opening the KB directly with the connector.
- Analyses each object with deterministic rules: type, structure, rules, events, variables, navigation, data access and references. No language model guesses at this stage: every conclusion is tied to the evidence that produced it.
- Relates the objects: it builds the dependency graph, the data model, the call trees, the data flow diagrams and the clusters (functional groups of objects that work together).
- Writes the documentation: one document per object, the data model, the clusters, a rebuild roadmap and the web viewer.
- Optionally, adds AI prose (the narrative): a language model writes explanations and the engine checks that every name it cites exists in the object before publishing them.
Read-only
UltraGX never modifies the KB. The connector opens the Knowledge Base read-only and has no write path at all; with an XPZ or a text export you do not even need GeneXus installed. Everything it generates is written to a separate output folder.
Editions
| Demo | Full | |
|---|---|---|
| Duration | 10 days from installation | per the licence |
| Inventory, graph, data model, clusters, call trees and diagrams | the whole KB | the whole KB |
| Detailed document per object (rules, pseudocode, implicit SQL, interface, original code) | 25 objects chosen by the engine | all |
| Knowledge packs, rebuild plans per cluster, AI narrative | sample only | all |
| Internet connection | required on every run (licence check) | not needed |
In the demo, the remaining objects get a summary document with their identity, location, profile, confidence, cluster and references, plus a notice that the detail is available in the full edition.
What data leaves your machine
- The KB analysis runs entirely on your computer. The code, rules and structure of your objects are not sent to any UltraGX server.
- Demo licence: when you register you send your contact details; on every run the licence service receives the licence identifier, the machine id, how many objects the KB has (counts only) and the names of the objects proposed for the sample.
- Full licence: there is no communication with the service after activation.
- AI narrative: if you use it, each object's document is sent to the provider you choose (Anthropic, OpenAI, Google, etc.) with your own API key. If you would rather nothing leaves your network, use a local model (Ollama or LM Studio). See AI narrative.
How to read this manual
- If you are trying the demo: Installation → First start and licence → Documenting a KB → Reading the result.
- If you already have a licence: Documenting a KB → AI narrative → User notes.
- If you are part of the migration team: Reading the result → Migrating with an AI agent → Command line.