Docbank

Follow one document through Docbank

This guide follows a single document from import to backup. Each step says what Docbank stores and what is left for you to decide. The processing steps apply only if you set up processing.

  1. 01 Import
  2. 02 Move and edit
  3. 03 Review processing
  4. 04 Extract text
  5. 05 Add embeddings
  6. 06 Find
  7. 07 Export
  8. 08 Manage outputs
  9. 09 Pick an interface
  10. 10 Back up

  1. Import the file

    Docbank copies the file into the vault and leaves the source where it was. The copy appears in a folder tree you can browse. Docbank records a checksum at import, so it can tell later whether the stored bytes have changed.

    Importing documents
    The web app showing a folder of imported documents in a synthetic vault
    A vault of synthetic documents after import.

  2. Move, rename, or replace it

    Every document has a node ID that stays the same for its whole life. Moving or renaming the document changes its path and nothing else. Replacing its content saves a new version, and the earlier versions you keep can still be opened or downloaded.

    Editing and versions
    The version history of one document, with a download button for the selected version
    Each saved version of a document can be downloaded.

  3. Review what processing will send

    Processing is set up as profiles. Choose one, and Docbank shows its plan for the selected version: each provider, where it runs, and the kind of content it will receive. Give consent once you have read the plan. A vault restored from a backup needs new consent before any network processing.

    Processing and consent
    A processing plan listing the providers and the content each one will receive
    The plan for one version, shown before anything is sent.

  4. Extract the text

    Processing produces a rendition: the document’s text as Markdown, saved against the version it came from. Start it from the web app, TUI, CLI, HTTP API, or Go API. The original file is not changed. With a supported provider, optical character recognition (OCR) reads text from scans and images.

    Document processing
    The extracted Markdown text of a saved document, shown in the web app
    The extracted text of one saved version.

  5. Add search by meaning

    An embedding is a list of numbers that stands for a piece of content, so similar content can be found by comparing numbers. Configure an embedding service, review what it will receive, and process the versions you want to search. The web app and TUI then offer Semantic and Hybrid search. Finding similar documents compares embeddings already in the vault and calls no provider.

    Search by meaning
    A list of documents ranked by similarity to the selected file
    Similar documents, ranked from stored embeddings.

  6. Find the document again

    Search file names and extracted text, then narrow the results by tag, folder, file type, or date. Save the queries and highlight terms you use often. A lexical (keyword) query can also be frozen: its rows, counts, and facets stay as they were when it ran, whatever changes in the vault afterwards.

    Searching
    A frozen query in the web app, with its result rows and facet counts
    A frozen query keeps the versions it found.

  7. Export the versions you reviewed

    Select documents, or take a whole frozen query. Docbank lists the originals and processing outputs the export will include, then builds a ZIP and verifies it. If a document changes after you start, the export still holds the version you selected. For search counts over a date range, download the CSV report with its evidence ZIP.

    Verified export bundles
    The export drawer reporting a verified ZIP that is ready to download
    A finished export, verified and ready to download.

  8. Processing outputs stay separate

    Text, previews, embeddings, and indexes are derivatives: outputs computed from a saved document. Docbank stores them apart from the original and records which version each one came from. Processing and retention rules decide when a derivative can be replaced or removed.

    Processing and retention
    A cycle of consent, processing, search, and removal around an original that does not change

  9. Use any interface

    The command line, web app, terminal browser, and HTTP API all go through the local daemon. A local agent connects with docbank mcp and works with the same versions and revision checks. A Go application can instead own a separate vault in its own process.

    Docbank for agents
    The CLI, web app, TUI, HTTP API, Go API, and agents all reading the same documents and versions

  10. Back up, then test a restore

    Create an incremental backup and verify it. Then restore it into a separate vault and check the result before you need it. A backup holds document content, metadata, and version history.

    Backup and restore
    Three steps: take a snapshot, verify it, and restore it into a separate vault

Next steps

Run the quickstart to try these steps on your own files. The task guides cover commands, automation, storage, and recovery.

Run the quickstart Browse the docs
Expanded documentation capture