Get started
Mizura overview
You ask a question. Mizura turns it into an analysis you can check: the data, the plan, the code, the checks, the results.
Mizura is a statistical analysis agent by Model Zero for Technological Solutions. You load a dataset and describe what you want in plain language, in English or Arabic. Model Zero, the model behind Mizura, plans the analysis and writes Python. The Python runs in a constrained sandbox on your data.
The point of the system is that it does not take the model's word for anything. Four things enforce that:
- Analyses follow methodology cards: written methods with assumptions, decision criteria, required outputs, and a runnable Python template. The library holds 83 cards in 12 families.
- Every number in an answer is checked against what the code actually produced. A number the code never printed goes back for correction. In a report, it blocks the document.
- Every run leaves an execution record with hashes of the code, the data, and the result. The replication package carries code, data, and records together, for replay and hash checks. Large dataset exports have additional input requirements; see Export & reproduce.
- Literature comes from Mizura's own search service. A report may cite only sources that were retrieved and opened.
Availability
Mizura is in development. Public installers are not yet available and subscriptions have not opened. These guides describe the current implementation for people with access to a development build. Check the Download page for release availability and Pricingfor published commercial terms. See Download & installbefore starting the quickstart.
Which version these docs cover
These guides cover the development source snapshot reviewed on, at product revision b25beab1 plus the reviewed working-tree changes. They are not instructions for a released installer. Your build may differ; compare its controls and errors with the guide before following a procedure.
Each how-to guide states what you need, what to do, and how to check the result. The quickstart is a guided lesson, the reference pages list controls and limits, and the concept pages explain the reasons behind the behavior.
What do you want to do?
| Task | Start here |
|---|---|
| Explore a new dataset | Common workflows |
| Choose and check an analysis | Best practices |
| Share reproducible work | Export and reproduce |
| Understand where data goes | Security and data flow |
| Fix a failed run | Troubleshooting |
| Customize the workspace | Settings |
Choose a path
Start with the quickstart if you are new. It takes one dataset through a full analysis. The concept pages explain how each part works. The reference collects what you need when something goes wrong.
Quickstart
Load a dataset, ask a question, approve a plan, review the output, and export a report. Five steps.
Download & install
Get the app for your platform, install it, launch Mizura.
Your data
Supported formats, importing from GitHub, Google Drive and Zenodo, inspecting a dataset before you trust it, size limits.
Workflow & plans
The workspace, sessions, plans and the Go button, steering a running turn, stopping it.
Methods
The 83 methodology cards, the 12 families they cover, and what a card can and cannot guarantee.
How code runs
Where Python runs on your device, what it cannot touch, the limits you will hit, and how R, Stata, SPSS and SAS code is handled.
Evidence & literature
Number grounding, flagged answers, literature search, and the citation rules for reports.
Reports & exports
Reports in six citation styles and five formats, section editing, notebooks, Quarto, replication packages.
Memory & conventions
What Mizura remembers about your decisions and datasets, how conventions are learned, and how to turn learning off.
Capabilities & reference
What Mizura can do, the Settings tabs, keyboard shortcuts, and fixes for common errors.
Who Mizura is for
Researchers with structured data. People who need a record of methods, code, and results that survives review by a colleague, a supervisor, or a journal. If that is not you, Mizura is the wrong tool. It is not a general-purpose chatbot, and it does not replace your judgment about study design, assumptions, or interpretation. Read its outputs with your own eyes.
What runs where
The engine runs on your computer. Your data, sessions, and outputs stay in a local data folder on your device, and analysis code runs locally in a sandbox. Prompts and analysis context go to Model Zero to generate responses. Retention and sharing are governed by the service privacy policy. See Security & data flowfor the boundary between local execution and remote services. Literature search contacts external services.
The privacy policy describes the product's retention, optional sharing, and feedback policy. The local engine cannot establish a remote provider's retention behavior. Review Security & data flowand the policy before using sensitive data.