Runtime reference
Novaber data and network boundaries
Novaber · Content reviewed
On this page
Novaber applications run in local containers. Accounts, authorization, downloads, updates and application-specific external APIs may still need a network connection. This page helps you distinguish those stages before use.
Check the application you choose
Applications can use different models, container images, file mounts and external services. Before processing sensitive material, review the application's documentation, configuration, input folders and output location.
This page explains product operation. It does not replace an application's data terms, service agreement or independent license.
Share only relevant information when reporting an issue
Share only relevant evidence after removing credentials and private content. Follow feedback guidance for reproduction steps and report details.
Where the network is involved
Visiting the website
Public product pages use website traffic analytics. Sign-in, account, payment and desktop authorization flows follow separate restrictions. Do not put passwords, tokens or private files in public feedback or links.
Sign-in and authorization
Email sign-in, account authorization and membership checks connect to account services. Local task execution does not mean the account system works entirely offline.
Installation and updates
The desktop installer is downloaded from Gitee. The app catalog and release information come from the website, and container images from their respective registries. Checking for and obtaining updates requires a connection.
Models and dependencies
An application's first run may download models or other dependencies. Their sources, caches and licenses depend on the selected application. Installing Docker alone does not make every model ready.
Running tasks
Applications run in local containers. Task authorization, usage or execution-result accounting may connect to account services. Network access by third-party models or external APIs depends on each application's documentation and configuration. Local execution does not make every stage offline.
Inputs and outputs
Each application manages its files and processing results. Check its mounted folders, output paths, external APIs and data documentation before use. The website does not make a universal promise that files never leave the device across all applications.
Before processing sensitive material
Use a small public sample to inspect import and export locations, model downloads and external-interface settings. This page is not a per-app network-traffic audit or a substitute for checking your configuration; it cannot establish that files never leave the machine or every stage works offline.
Where are setup skills and diagnostic resources read from?
The documentation site’s current setup entry and pinned Skill 1.0.3 entry provide text resources. Keep references, the diagnostic script and checksum manifest from the same package. Reading a page neither installs software nor runs scripts; verify the downloaded script’s hash and review its contents first.
Skill 1.0.3 checks NVIDIA driver major >= 575 and Windows driver >= 576.57 for the project CUDA 12.9.1 baseline. Its diagnostic script reads local evidence without automatically uploading the report. Desktop, driver, WSL, Docker, app image and model installation can still access their respective release or download services. See runtime requirements for driver and container checks.