Is my data used to train models?
No. Uploads, artifacts and chats are used only to run your requests. Managed models are called through provider APIs that don't train on the data; connector data follows the same rule. See the privacy policy.
Can I keep everything on my machine?
Not yet. On the web, uploads are stored in your private workspace and used only to run your requests (see the privacy policy). The desktop app, being prepared for release, will read files and run analysis code locally so that only observations go to the server. It will also be able to run a whole turn on a local model (Customize → Local AI): the model will run on your computer, prompts and tool observations will pass through Neutropic servers in memory to reach it, but no third-party AI provider will receive your data, and local turns will never use managed tokens — on every plan, including Free. Your own API keys will come with it and will be free on every plan too. See Effort levels & models.
How accurate are the citations?
Every [n] in a report points to a source in the screened corpus; the reviewer checks that the cited source supports the claim. Sources are real database records, not generated. Still read the sources you rely on — the reviewer catches unsupported claims, not scientific disagreement.
Which statistics can it run?
Descriptives, correlation, t-tests, ANOVA, regression (linear/logistic), mixed models, Bayesian estimation, reliability, EFA/CFA/SEM, IRT, power analysis, non-parametric alternatives, multiple-comparison correction, effect sizes with CIs — plus domain instruments (SUS, NASA-TLX, UEQ, SD, conjoint) and biosignal pipelines. Full list: Method skills.
Can I edit what it writes?
Yes — reports and manuscripts are editable in place, figures open in the vector editor, and every change is versioned. See Artifacts & the panel.
What does a turn cost?
At Basic effort a literature review uses a few hundred thousand managed tokens; statistics and figures use less; Deep and heavier models use more. The Free plan's ≈ 3M managed tokens a month cover a handful of full research turns. See Effort levels & models and Plans & tokens.
Which languages are supported?
English, Korean, Japanese and Chinese for the interface, the agent's answers and the artifacts. Literature search runs against international databases regardless. See Language settings.
Can my lab share sessions?
You can send anyone a read-only link to a chat — they see the conversation and its artifacts without an account (see Share a chat). Working on a chat together, with shared projects, skills and token pools, is planned for team plans — see the changelog.
Where do I report a bug?
Get help with the session link.