Document sources

Colrows answers questions from your documents. Connect a document source, and Colrows parses its files, indexes them for search, and keeps them in sync. Documents belong to your organization: once a file is synced, every member can get answers from it. This page covers how to connect sources, which formats are read, and how sync works. To ask questions, see Ask your documents; for who can see and manage what, see Document access.

Turn on document processing

An administrator enables documents once per organization in Administration → Settings → General with the Enable Document Processing switch. You can add and test connections while it is off, but syncing documents requires it to be on.

Connect a source

Open Administration → Connections → Documents and add a connection. Administrators and curators add and configure connections. All connectors are read only: Colrows reads documents and never writes back. It does not read the source's sharing settings, so connect only the files your whole organization may use.

Folder sources list up to 20 levels of subfolders and 5,000 folders by default. A scan that reaches a limit, or cannot list a subfolder, is reported as incomplete, and files it did not reach are not removed. Credentials are stored in your secret manager - Colrows keeps only a reference and never returns the secret to the browser. Use Test Connection before saving.

Each connection can also filter files by extension, file name pattern, created and modified dates, and file size. Files that stop matching are removed from search on the next sync.

Knowledge uploads

Consensus → Document Upload is a different path. A file uploaded there is a semantic knowledge source: Colrows extracts business terms, metrics, and other knowledge from it as signals for curators to review. Knowledge uploads are never searched, cited, or used to answer document questions. Administrators and curators upload them, with optional Related databases and a Description; the list shows each upload's knowledge cycle.

Formats and parsing

• Supported formats: PDF, DOCX, Markdown, and plain text, up to 32 MiB and 500 pages per file.
• Google Docs, Sheets, and Slides are read by exporting them as PDF (subject to Google's export limits).
• Parsing runs locally in a sandboxed process. Tables keep their headers, footnotes stay linked to the text that references them, and pages with no readable text are reported rather than dropped.
• Scanned pages are read with OCR (Tesseract, English). Image and chart contents are not interpreted.

Sync

• On demand: open Docent → Ingestion Pipelines, select the connection, and choose Sync Documents. The scan lists the source and queues new or changed files; you can also retry files that failed.
• Scheduled: add a sync automation on the connection's Automations tab. Without one, a connection syncs only on demand.
• Incremental: unchanged files are skipped by S3 version ID, Google Drive revision, or OneDrive eTag.
• Deletions are applied only after a complete scan confirms a file is gone, so an outage or a transient error never removes documents.
• Unconfirmed files are hidden: a file whose state at the source a sync cannot confirm is kept out of answers until a later sync confirms it.

Processing status

Docent → Ingestion Pipelines tracks each connection's files from discovery to the search index, with Overview, Automations, Sync History, and Files tabs. A file is Pending, Processing, Indexed, Failed, Source Access Unresolved, Excluded, or Superseded, and is answerable once it is Indexed. A failure shows a reason, and a new version of a document is swapped in only after it finishes processing, so questions always run against the last good version.