LegacyExodus is building a verification-driven software modernization platform. Today’s deterministic JS/TS foundation turns source structure and program flow into inspectable evidence. Future stages govern code transformation and independent verification.
LegacyExodus is designed as a continuous, four-stage pipeline that advances from raw repository discovery to independently verified target code. We separate what is currently verified from what remains active roadmap architecture.
Inventory the repository without assumptions. Classify source files, identify workspace boundaries, resolve module topologies, and construct deterministic dependency maps before any code transformation begins.
Eliminates blind spots. Engineering leaders gain immediate, factual visibility into codebase scale, module coupling, and migration complexity before committing budget.
Extract ground-truth structural intelligence from code. Parse ASTs, model lexical scopes and symbols, reconstruct interprocedural call graphs, build control-flow and data-flow graphs, and compute conservative taint facts.
Recovers lost business logic and data propagation paths without runtime execution risk. Developers understand exact dependencies and branch reachability before touching legacy code.
Transform verified evidence into modern code through bounded AI reasoning. The AI gateway receives immutable, source-grounded graph snapshots to generate candidate architectures and typed target code (Rust/Axum/SQLx and modern TypeScript).
Replaces unguided, hallucination-prone AI code generation with disciplined, evidence-bounded transformation. Every proposed line references verified source semantics.
Independent verification and human governance. Evaluate candidate migrations in isolated compiler sandboxes, run differential tests against original behavior, enforce fail-closed verification gates, and require human architectural sign-off.
Zero unverified code enters production. Independent validation gates ensure the transformed system compiles, passes differential tests, and receives human authorization before deployment.
Select a topic, inspect its relationships, and see the boundary between the current foundation and future architecture.
Source repositoryJavaScript and TypeScript are the current analysis focus. This repository layout is schematic.
VyrixScout inventories files and reconstructs modules and dependencies. Discovery establishes the system boundary before deeper analysis.
classifyScoreRepresentative function name from the public sample; this AST is an explanatory schematic.
NolaraStruct models syntax, scopes, and symbols. Stable structure makes program relationships inspectable instead of leaving them buried in text.
Function entryEntry into classifyScore. This is static control flow, not an execution trace.
Control flow exposes possible paths. Data flow concerns how values relate across program operations. The public sample demonstrates selected intraprocedural CFG and reachability; DFG belongs to the broader documented foundation.
score parameter>= 70 comparisonIllustrative value dependency, not captured DFG output. The public CFG sample excludes data-flow analysis.
Source revisionArtifacts need source provenance to remain interpretable.
Canonical artifacts retain relationships and source provenance. Reproducibility lets engineers compare and revisit a model; it does not guarantee a complete model or behavioral equivalence.
Evidence contextVersioned source evidence supplies the intended reasoning boundary.
A bounded reasoning layer is intended to explain evidence and propose modernization work. Candidate changes would require independent checks and human decisions. No live AI integration is provided here.
Source languageJS/TS is the current foundation. Java, PHP, Python, and Ruby are exploratory directions.
Language-aware adapters and explicit semantic mappings are intended to connect source and target ecosystems. JavaScript/TypeScript is the current focus; additional adapters and target generation remain future work.
Documented capabilities in the engineering foundation, with explicit boundaries.
A repository contains more than its directory tree. Module relationships and resolution rules determine how its parts connect.
Text search cannot reliably explain which declaration a name refers to or how a function is called.
Branches and values move across functions. Understanding those paths is necessary before assessing a change.
Understanding is hard to revisit when analysis results are transient or dependent on an unrecorded run.
Deterministic evidence defines what is known. AI helps reason about what comes next. Independent verification determines what can be accepted.
We reject the notion that raw LLMs should be given unconstrained read/write access to production codebases. In LegacyExodus, generative reasoning operates within six rigorous governing boundaries:
Platform Vision: An end-to-end governed migration platform that discovers, understands, transforms, and validates complex enterprise codebases across multiple languages.
Available Engineering Foundation: The active, high-performance JavaScript and TypeScript static-analysis CLI. It performs repository inventory (VyrixScout), AST normalization, scope/symbol modeling, semantic call graphs, CFGs, DFGs, and conservative taint facts (NolaraStruct), serializing to streaming JSON, memory, and SQLite.
Roadmap: Automated transformation planning (ZelvoxForge), isolated compiler sandboxing and differential equivalence testing (VymosGate), and language adapters for PHP, Ruby, and Java.
This website does not provide a hosted dashboard or accept proprietary code uploads. Capability status is based on our engineering repository audit as of October 2026.
The intended AI gateway supports architecture explanations, risk analysis, transformation strategies, and candidate changes. Source evidence constrains the proposal; verification and human review determine acceptance.
Versioned source context
Typed inputs & limited tools
Explicit transformation scope
Proposed target-specific code
Independent checks & results
Approve, revise, or reject
AI proposes. Independent checks supply evidence. People decide. These future controls require implementation and evaluation.
The goal is language-extensible modernization, including cross-language migration where mappings can be established. Additional adapters, intermediate representations, and target generators remain future work.
Repository discovery & semantic analysis
No shipped adapters or committed sequence.
Language-aware adapters, structured knowledge & IR, explicit mappings and transformation contracts.
Target-specific generation and independent verification.
Rust is an intended target. Other destinations require their own semantic mappings and evidence.
Follow the engineering, explore the approach, or get in touch.