The process itself is the audit trail.
AXIO Reader executes process models deterministically. The evidence isn't produced alongside the run — it is the run. No logging after the fact, no reconstruction.
Local · No account · No server round-trip
Quick start guide (PDF) ↓ · Example models to try ↓
31 seconds. No words. ♪ with sound
Why this matters now.
AI agents work non-deterministically
An agent loop is strong where the path to the answer isn't known up front. The same mechanism makes the run hard to verify: two runs on identical input can take different paths.
What happened can be reconstructed afterwards — but not readily repeated. For regulated processes, that is precisely the problem.
Regulation demands traceability
ISO 42001, the EU AI Act and NIS2 all demand the same thing: that a process be traceable and reproducible. The EU AI Act attaches substantial administrative fines to non-compliance.
AXIO is the deterministic alternative: same input, same run, same verifiable trail — repeatable at every audit.
One model. Read three ways.
A model is a data structure — described, executed, audited. The same structure, read three ways. It cannot drift, because it only exists once.
Described
You load a model — declarative, in plain text, without implementation detail. What should happen stands there as structure.
Steps: defined
Human approval: provided for
Audit trail: active
Executed
AXIO executes the model deterministically. No guessing, no hallucinating. Where a human approval is provided for, the run waits.
▸ Step by step
⏸ Waiting for approval…
▸ Continuation
Audited
Every step, every value, every decision sits in the run. The run itself is the trail — reproducible, without a separate log.
Every step: recorded
Reproducible: yes
Separate log: not needed
Models to try.
AXIO Reader already ships with a starter model. Download the examples below in addition — open them in the Reader, execute them, compare the trail.
Loading models…
Where it holds up.
Anywhere "it was configured that way once" isn't an answer.
Compliance
ISO 42001
Assessment steps as an executable model — traceable, not a PDF checklist.
Regulation
EU AI Act
Processes with a complete record — auditable from the start, reproducible at every review.
Quality & GRC
Internal audits
Review processes that document themselves. The evidence is not an afterthought.
AI governance
Making LLM workflows verifiable
The deterministic core stays verifiable; the non-deterministic part is enclosed as its own step and held in the run.
Honest answers.
Is AXIO certified?
No. AXIO is aligned with ISO 42001, the EU AI Act and NIS2, and designed to deliver traceability — that is, to enable compliance. The tool itself carries no certification, and we claim none. Certifying a process remains the task of the relevant assessment body; AXIO provides the verifiable basis for it.
Is there AI inside AXIO?
The execution core contains no AI model — which is why it is deterministic. AXIO is not AI-free, it is AI-agnostic: where a non-deterministic step makes sense, it can be enclosed in a controlled way, visible as its own step and held in the run.
What does Reader cost?
Nothing. AXIO Reader is the first public release and it's free. It runs locally, without an account and without a server round-trip. AXIO Studio, the full authoring environment for building your own models, follows later as a paid version.
Do I need a model to get started?
No. AXIO Reader ships with a simple example model — hello-axio — automatically. It lets you follow the core straight away: load, execute, produce the same run again, compare. Further models follow as downloads — deliberately step by step, developed with the people who use the Reader. This is only the beginning.
Does AXIO run on my system?
The Reader is available for macOS (Apple Silicon) and Windows (x64), each as a local desktop application. We'll provide a Linux build once there is enough interest — write to us if you need one.