The Analytical Compiler for Neural Architectures
NEURAX predicts the cost, memory, and performance of neural network architectures before training - in under 50 ms, with zero GPU, and fully deterministically.
Live Demo · Documentation · API Reference · Releases · Contributing
NEURAX is an analytical compiler for neural network architectures. Whereas training frameworks (PyTorch, TensorFlow) execute models and runtime compilers (IREE, OpenXLA) lower them for execution, NEURAX operates at design time: it answers the questions you need resolved before committing GPU resources.
- Will this architecture fit in VRAM?
- What is the training cost on 8x H100?
- Where are the memory bottlenecks?
- Is inference stable? What is the hallucination risk?
- Which parallelism strategy is optimal?
All in under 50 ms. Zero GPU required. Fully deterministic.
- 11 architecture families - Transformer, CNN, MoE, SSM, Diffusion, GNN, GAN, RL, SNN, RNN, Experimental.
- 680+ configurable blocks - Attention, MLP, Conv, Embedding, Normalization, and more.
- 88 reference templates - From GPT-4 to Stable Diffusion, production-ready architectures.
- <50 ms analysis - Full 10-pass IR pipeline on 8B-parameter models.
- 55+ metrics - FLOPs, VRAM, latency, cost, energy, carbon emissions.
- Deterministic - Identical input always produces identical output.
- No GPU needed - Pure analytical formulas; runs in the browser or CLI.
- Drag-and-drop architecture builder with 680+ blocks.
- Real-time validation of connections and parameters.
- Parameter editing directly on the canvas.
- Export to 7 formats - PyTorch, ONNX, Triton, MLIR, Rust/Burn, JSON, Network Graph.
- Natural-language design - "Create a transformer for image classification".
- Multi-provider support - OpenAI, Anthropic, Google, Mistral (BYOK).
- Auto-validation of topology with optimization suggestions.
- Fully private - your API key never leaves the browser.
- 22 configurable parameters - sampling, context, model behavior, stress testing.
- 10 analytical widgets - stability, entropy, hallucination risk, attention focus.
- Predict before serving - know if your model will behave before deployment.
- Multi-year cost, carbon, and scaling projections (3-5 years).
- Regulatory compliance - EU AI Act, CSRD, DSA tracking.
- Hardware migration planning with data.
NEURAX operates like a traditional compiler, but for neural architectures:
flowchart LR
A[Design Architecture] --> B[Analytical Compilation]
B --> C[Engineering Report]
B --> D[MLIR Code]
C --> E[Cost Predictions]
C --> F[Memory Analysis]
C --> G[Performance Metrics]
D --> H[LLVM IR]
H --> I[CPU/GPU Execution]
graph LR
Input[model.json] --> P1[1. Architecture IR]
P1 --> P2[2. Graph IR]
P2 --> P3[3. Tensor IR]
P3 --> P4[4. Operator IR]
P4 --> P5[5. Compute IR]
P5 --> P6[6. Memory IR]
P6 --> P7[7. Parallelism IR]
P7 --> P8[8. Hardware IR]
P8 --> P9[9. Cost IR]
P9 --> P10[10. Report IR]
P10 --> Output[report.json]
Each pass transforms the representation and computes specific metrics:
| Pass | Computed metrics |
|---|---|
| Architecture | Layer count, model type, global parameters |
| Graph | Topology validation, DAG structure, fan-in/fan-out |
| Tensor | Shape inference, dimension resolution, memory layout |
| Operator | FLOPs per op, parameter count, operation types |
| Compute | Total FLOPs, throughput, backward/optimizer overhead |
| Memory | Peak VRAM, activation/gradient memory, fragmentation |
| Parallelism | Tensor/pipeline/expert parallelism, efficiency scores |
| Hardware | GPU utilization, bandwidth, ridge point, latency |
| Cost | Training cost (USD), time (hours), energy (kWh), CO2 (kg) |
| Report | Consolidated metrics, diagnostics, recommendations |
NEURAX is a full-stack platform with 5 integrated surfaces:
graph TB
subgraph Frontend["Frontend Layer"]
UI[Web UI - React 18 + TypeScript]
CLI[CLI - Rust binary]
TUI[TUI - Ratatui terminal]
end
subgraph Services["Service Layer"]
API[HTTP API - Actix-Web, 38 routes]
Agent[AI Agent - FastAPI + LangChain]
MCP[MCP Server]
end
subgraph Engine["Analytical Engine"]
Parser[neurax-parser]
IR[neurax-ir - 10 passes]
Core[neurax-core - orchestrator]
Formulas[neurax-formulas]
HW[neurax-hardware-db]
MLIR[neurax-mlir - 13 dialects]
end
CLI --> Core
TUI --> Core
API --> Core
Core --> IR
Parser --> IR
Formulas --> IR
HW --> IR
Core --> MLIR
| Component | Language | Purpose |
|---|---|---|
| neurax-ui | React 18 + TypeScript | Visual canvas, metrics dashboard, AI chat |
| neurax-service | Rust (actix-web) | REST API, SSE streaming, auth, billing |
| neurax-agent | Python (FastAPI) | LangChain-powered architecture planning |
| neurax-core | Rust | Pipeline orchestrator, ONNX export |
| neurax-ir | Rust | 10-dialect analytical IR |
| neurax-mlir | Rust + MLIR | 13 custom dialects, LLVM 18 backend |
| neurax-parser | Rust | JSON schema to strongly-typed AST |
| neurax-formulas | Rust | Per-architecture analytical formulas |
| neurax-hardware-db | Rust | GPU/CPU specs (20 GPUs, 2 CPUs) |
| neurax-cli | Rust | Command-line interface |
| neurax-tui | Rust (Ratatui) | Terminal user interface |
| neurax-mcp | Python | Model Context Protocol server |
.
├── neurax-core/ # Pipeline orchestrator, ONNX export
├── neurax-ir/ # 10-dialect analytical IR
├── neurax-mlir/ # 13 custom dialects, LLVM 18 backend
├── neurax-parser/ # JSON to strongly-typed AST
├── neurax-formulas/ # Analytical formulas
├── neurax-hardware-db/ # GPU/CPU spec database
├── neurax-cli/ # Command-line interface
├── neurax-tui/ # Terminal UI
├── neurax-service/ # Actix-web HTTP API
├── neurax-agent/ # Python AI planning agent
├── neurax-mcp/ # MCP server
├── neurax-ui/ # React web frontend
├── docs/ # Project documentation
├── examples/models/ # Reference architecture configs
└── .github/workflows/ # CI (LLVM 18 / MLIR build)
git clone https://github.com/rustnew/NEURAX.git
cd NEURAX
./start-dev.sh
# Web UI -> http://localhost:8081
# API -> http://localhost:9098
# Agent -> http://localhost:8099cargo build -p neurax-cli --release
./target/release/neurax analyze models/gpt2_small.jsondocker compose up -d
# Access at http://localhost:8081NEURAX ships with 88 reference templates across 11 families:
| Family | Examples |
|---|---|
| Transformer / LLM | GPT-2, LLaMA 2/3, BERT, Mistral 7B, Falcon 7B |
| Mixture-of-Experts | Mixtral, DeepSeek MoE, Qwen2-MoE, DBRX |
| CNN / Vision | ResNet, VGG, EfficientNet, MobileNetV2, ConvNeXt |
| State-Space Models | Mamba, Mamba2, ViM |
| Diffusion | DDPM, Stable Diffusion, Imagen, DALL-E 3, FLUX |
| GNN | GCN, GAT, GIN, GraphSAGE |
| GAN | DCGAN, StyleGAN, ProGAN, CycleGAN |
| Reinforcement Learning | DQN, PPO, SAC, A2C, TD3 |
| Spiking Neural Networks | LIF SNN, Spiking ResNet, Spikformer |
| RNN / LSTM / GRU | BiLSTM, LSTM Seq2Seq, GRU Seq2Seq |
| Experimental | Neural ODE, Liquid Time-Constant, Quantum Hybrid |
| Document | Description |
|---|---|
| Architecture & Design | System architecture, data flow, design principles |
| API Reference | 38 REST endpoints, auth, schemas |
| Deployment Guide | Production and Docker deployment |
| Roadmap v2.0 | Development roadmap |
| Contributing | Development workflow and code style |
| CHANGELOG | Version history |
| Security | Security policy and vulnerability reporting |
- 10-pass analytical IR pipeline
- MLIR compiler backend (13 dialects)
- Visual canvas with 680+ blocks
- AI copilot agent (multi-provider)
- Inference Intelligence (22 parameters)
- Time Machine (multi-year projections)
- Multimodal (VLM) model support
- Modern landing page and avatar system
- NEURAX-MLIR to IREE kernel lowering
- Public benchmark suite (predictions vs measured)
- Batch hyperparameter optimization API
- PostgreSQL for project persistence
- Distributed training projections
- Model hub with HuggingFace integration
- Fine-tuning cost projections (LoRA, QLoRA)
- Kubernetes production deployment
- Collaborative multi-user editing (CRDT)
NEURAX follows Semantic Versioning. Releases are published on the Releases page and documented in the CHANGELOG.
Contributions are welcome. See CONTRIBUTING.md for the development workflow, project layout, code style, and how to open a pull request. Please read the Code of Conduct.
NEURAX is free and open source, built and maintained by the community. Your sponsorship helps us keep the project sustainable and growing.
Why sponsor NEURAX?
- Support the development of the first analytical compiler for neural architectures
- Help democratize ML architecture design and save GPU costs
- Get your logo featured here and in our documentation
Sponsorship tiers:
- $5/mo - Thank you + name in our sponsors list
- $25/mo - Logo in the README + early access to new features
- $100/mo - Priority support + case study feature
- $500/mo - Monthly consultation + landing page logo
Every contribution, no matter the size, makes a difference. Thank you for supporting open source! 🙏
Funding applications:
Community & growth:
- Go-To-Market Strategy
- ArXiv Technical Report
- ArXiv Endorsement Outreach
- Zenodo Deposit Guide
- Promotion Checklist
- Social Media Assets
NEURAX is open-source software licensed under the MIT License. See LICENSE for the full text.
NEURAX builds on the shoulders of giants:
- MLIR - Multi-Level Intermediate Representation framework
- LLVM - Compiler infrastructure
- Rust - Systems programming language
- React - UI framework
- shadcn/ui - Component library
Built by Fossouo.