Skip to content
OrbitLab 0.1.7-beta · Commercial Beta

Research software for people who do the research

OrbitLab takes a raw dataset to a defensible statistical analysis, publication-quality figures and a literature review — computed locally, on your own machine.

Windows 10 & 11 · 64-bit · 81.2 MB · Released 3 Aug 2026

OrbitLab by StarshipAI

OverviewProposalData ExtractionStatisticsResultsManuscript

Analyses

6

Assumptions

Checked

Engine

Local

Independent-samples comparison

Welch's t-test · two-tailed

Deterministic · No AI
nmean diff95% CIpeffect size
Interface illustration of the OrbitLab workspace — no participant data shown.

One workspace

Three modules, one continuous workflow

Each module hands its output to the next. Results computed in Research Lab are the same results Charts Studio draws and the same numbers your manuscript quotes.

Stable

Research Lab

Plan a study, build an extraction sheet, validate it, then run the analysis the design actually calls for. Descriptives, t-tests, ANOVA, Mann-Whitney, Kruskal-Wallis, correlation and 2×2 measures with confidence intervals.

Beta

Charts Studio

Publication-quality figures generated straight from saved results, so the numbers in the figure and the numbers in the table cannot drift apart.

Beta

Literature Search

One search across PubMed, OpenAlex, Crossref and Semantic Scholar, with duplicate records merged into a single ranked result.

How it works

From question to write-up without leaving the app

  1. 01

    Describe the study

    State your question and design. OrbitLab proposes the variables you need to collect and the analysis that fits.

  2. 02

    Build the extraction sheet

    Import a spreadsheet or enter data directly. Separators, line endings and thousands separators are handled and flagged rather than misread.

  3. 03

    Validate before you analyse

    Missing values, type conflicts and plan mismatches are resolved up front, so the analysis runs on data you trust.

  4. 04

    Compute, chart and write up

    Run the recommended tests, generate figures from those exact results, and export to Word or PDF with the numbers already in place.

Included with every plan

A research assistant that already knows OrbitLab

Sign in with your OrbitLab account and just ask. It answers questions about the software, works out which statistical test your design actually calls for, reads your output with you, and coaches you through every section of the manuscript — abstract, introduction, methods, results and discussion.

It runs in the browser, so you can use it on a phone or a second computer while OrbitLab stays installed on your one licensed machine. Signing in here does not use up that machine.

Free with an active plan. There is nothing extra to buy.

Things people actually ask it

  • Which test should I use for two independent groups?
  • My data are not normally distributed — what now?
  • Here is my abstract. Cut it to 250 words.
  • I cannot articulate the gap in my introduction.
  • My p-value is 0.07. What do I write?
  • Walk me through my first project, step by step.

It replies in the language you write in, and keeps anything destined for your manuscript in English.

It knows the product

Answers about OrbitLab come from its documentation, and it tells you when it is not certain instead of inventing a menu you will never find.

Your conversations stay in your browser

Chats are never stored on a StarshipAI server. Only a per-day message count is kept, so your allowance can be enforced.

It will not make things up

No invented citations, no invented numbers, and no methods section describing work nobody did. It edits and coaches; it does not fabricate.

Why it is built this way

Trust is a feature, not a footnote

Statistical software is only useful if you can defend what it produced. These are the constraints OrbitLab holds itself to.

Your data never leaves the device

Research projects, datasets and participant rows are stored locally. Statistics are computed on your machine, not in a cloud.

Deterministic where it counts

Every inferential result is computed by a deterministic engine — no model in the path. Run it twice and you get the same number.

Honest about its limits

A smaller-than-planned sample is reported as a limitation to write up, not silently swallowed. Conflicts are surfaced, not hidden.

Updates that respect your work

Signed release metadata, SHA-256 on every installer, and an update prompt that tells you what changed before you accept it.

0.1.7-beta
Current release
Published 3 Aug 2026
100%
Local computation
No dataset leaves your device
8
Public builds shipped
Every one still downloadable

Start with your next dataset

Download the current Commercial Beta and run a full analysis end to end. Your projects stay on your machine.