NOVABERDocs

Loading search index…

    Documentation menu

    Runtime reference

    Import documentation into Obsidian and AI knowledge tools

    Novaber · Content reviewed

    On this page

    Novaber Docs maintains Markdown and knowledge indexes in an independent Gitee repository. Clone the repository, then use Git to receive changed files. The website does not build or host a knowledge ZIP. The repository is currently private and requires access permission.

    Clone and open the documentation

    git clone --branch develop https://gitee.com/novaber/novaber-docs.git
    cd novaber-docs
    

    Start with knowledge/notes/index.md and choose the Chinese or English index. Articles in knowledge/notes/ already synthesize instructions by task, with relative Markdown links, local images and sources.

    In Obsidian, open knowledge/notes/ as a vault; no additional plugin is required. Keep personal notes separately instead of editing generated content managed by Git. Original installation references are under sources/; opening the vault does not execute included scripts. Installers, current official execution instructions and account services still need network access.

    Receive updates

    git pull --ff-only
    

    This updates the cloned repository rather than downloading a new full archive. Save personal work first. If you changed managed files, inspect the differences before merging updates. Maintain official articles in their source directories and regenerate the knowledge directory through the documented maintainer workflow.

    Connect retrieval or a model

    • knowledge/manifest.json: document IDs, language, product versions, Markdown paths, source hashes and review information.
    • knowledge/chunks.jsonl: content split by headings and blocks, with citations, product versions and hashes.
    • knowledge/releases.json: article product/version metadata for filtering, rather than a separate release-record section.
    • knowledge/documents/: complete attributable Markdown. Use knowledge/notes/ for offline reading and relative links.

    Retain at least document_id, language, product_id, product_version, is_current, source_sha256 and citation_url. Filter by language and product, excluding explicitly historical records for current usage questions. Retrieve a requested version when needed and retain its article and heading citation.

    Chunks do not imply compatibility with every model’s context window. The receiving system must handle oversized blocks and build its own retrieval or vector index. This project does not configure model accounts or upload documents automatically.

    Keep versions and provenance when updating

    Compare the manifest revision and record_sha256, replace chunks for changed articles, and remove articles deleted from the new manifest. Desktop, app and setup-skill versions are independent. Source hashes and note export hashes are separate because local links and image paths are converted.

    The committed knowledge directory excludes unreleased local previews, internal maintenance logs and private configuration. Read the current official setup skill for installation; historical references do not replace current execution instructions.

    How should retrieval handle setup skill versions?

    The setup skill’s product ID is novaber-setup; the current documentation uses Skill 1.0.3. Answers about NVIDIA GPU preparation should retain driver major >= 575, the project CUDA 12.9.1 Windows driver baseline >= 576.57, and the requirement to verify the host, container and app separately. See runtime requirements for the complete decision rules.

    Pin the skill version for an installation or resumption, including its references, script and checksum manifest. An old report does not establish current state, and desktop, app and skill versions must remain separate. The current skill entry follows the current version selection; Skill 1.0.3 supports a fixed-version review.

    Sources