Auditable · Traceable · No black boxes
MatrixAI Studio

Build a neural network
you can audit

Describe your model in plain language. MatrixAI builds, trains and deploys it with cryptographic auditing at every step. No PyTorch exposed. No hidden dependencies.

What that means — and what it does not
Try it online ↗What you get
PricingSee how it works ↗Cases you can redo ↗Use cases ↓
13,413 tests passing
ONNX + WASM
GPU optional

One tool, two ways in

MatrixAI full pipeline: Prompt → .semantic → .mxai IR → .mxtrain → ParameterSet → HTTP API. Features: Sandbox, Cryptographic hash, Optional GPU, Edge bundle, ONNX / WASM
SandboxCryptographic hashOptional GPUEdge bundleONNX / WASMHugging Face
01

A prompt — or your CSV

Describe the problem in plain language, or upload a CSV: MatrixAI infers the schema, proposes the target column and lets you correct types and ranges before anything is built.
02

Supervised proposal

MatrixAI proposes a model. You review, adjust or approve it. Nothing is built without human oversight.
03

.mxai model — audited IR

The model compiles to a human-readable intermediate representation. Every layer has a name, type and hash.
04

Traceable training

Versioned parameters, recorded metrics, gradient traces. Every training run is fully reproducible.
05

Serve, export or publish

Served over HTTP with Bearer auth, exported to ONNX, WASM or a self-contained bundle that runs without MatrixAI — or published straight to Hugging Face.

Capabilities

Works onTabular dataTime seriesFree text — transformer models
Transformer models train on torch, which ships in the GPU package.
Auditability

Cryptographic hash on every artifact

Every model, ParameterSet and training trace carries a SHA-256. You know exactly which model version made each decision, when and on what data.
✓ model_hash: sha256:4af2…c391
✓ param_set_id: ps-20260519-001
✓ trace_signature: hmac-sha256:…
Declarative

You define the model, not the math

The .mxai format is human-readable. Every layer has a name, type and declared parameters. MatrixAI handles the tensor arithmetic.
NETWORK CreditScorer {
  LAYER Dense units=64 act=relu
  LAYER LayerNorm
  LAYER Dropout rate=0.1
  OUTPUT Probability
}
— no PyTorch exposed
Safe actions

Explicit contract before acting

Real actions (email, HTTP, database) require a signed .mxact file, a successful dry-run and a declared rollback. Nothing executes without a trace.
Portability

ONNX, WASM, Edge bundle, GPU

Export the trained model to ONNX for universal inference, WASM for the browser, or a self-contained edge bundle for production. Optional GPU without changing a line of code.
ONNXWASMEdge bundleHTTP APIGPU opt.

Pricing

The neural network engine is free. The Studio package is €2.

The core engine is open source and always will be. The Studio package removes all setup friction so your team is productive from minute one.

MatrixAI Core

Free
Open source · AGPL v3
  • ✓Full engine · production ready
  • ✓CLI interface
  • ✓REST API + Swagger
  • ✓ONNX / WASM export
  • ✓GPU optional · Docker ready
  • ✓AGPL v3 — fully open source
View on GitHub ↗

MatrixAI Studio

Recommended
€2one-time
Pre-configured · No setup · Ready in minutes
  • ↳Everything in Core
  • ✓Studio pre-configured out of the box
  • ✓One-click Docker Compose setup
  • ✓CPU and GPU packages, image included
  • ✓Licence for 3 machines · 3 downloads
  • ✓Two interfaces included: classic and phases (Workbench in progress)
No account needed · credentials sent by email · Stripe
14-day refund — full money back within 14 days, unless you have already started your first download. You accept that when you download, and your licence stays valid meanwhile.
The core engine will always be free and open source. · The €2 funds continued development.

Use cases

Financial

Regulatory traceability
The problem
The regulator asks which model approved which loan. Today the answer is a spreadsheet and a prayer.
The solution
MatrixAI records the model hash, parameters and decision trace for every prediction. The audit response is a signed, timestamped file.
↗ Every credit decision is auditable by design

Healthcare

Clinical explainability
The problem
The model predicts high risk. The committee asks why. The black box has no answer.
The solution
Every prediction includes the human-readable explanation generated by MatrixAI Studio. Defensible in front of any medical committee.
↗ Every risk decision carries a readable justification

Operations & Support

Intelligent routing
The problem
The support team manually classifies and routes tickets. It does not scale. Errors cost hours.
The solution
A composite model with embeddings categorizes and routes tickets automatically. No black box: each assignment explains its reasoning.
↗ Manual routing eliminated without losing explainability

Operational agents

Audited actions
The problem
Automated agents act without an explicit contract. When something fails, there is no trace and no rollback.
The solution
MatrixAI executes real actions under an HMAC-SHA256 signed .mxact contract, with a mandatory dry-run and declared rollback.
↗ Automated action + signature + rollback = responsible operation

Tests and traceability

13,413
Tests passing
✓
Decisions traceable
0
Failing
6,179
Core engine
4,925
Studio
214
Real browser
2,095
Product backend
Automated test suites — 13,413 tests0 failed