knowmindknowmind
DokumentationQuickstartAnbindungReferenzFehlerbehebung
DEEN
knowmindknowmind
DocumentationQuickstartIntegrationsReferenceTroubleshooting
DEEN

Tasks

Upload documents

Longer texts and PDF can be pushed directly into knowmind. knowmind splits the document into sensible sections, generates a vector embedding per section, and makes the content semantically searchable.

Audience
All plans. Recommended once you start storing structured sources such as handbooks, contracts or whitepapers in your Memory Store.

Prerequisites

  • Active knowmind account
  • Document in Markdown, plain text or PDF (text-based PDF; scanned PDFs without OCR will not work reliably)
  • Maximum size per document: 10 MB
  • For the CLI option: knowmind CLI installed and signed in
Note

When a document, when a memory?

Memories are short (a note, a memo). Documents are longer (a handbook chapter, a contract, a specification). knowmind chunks documents automatically; a memory stays as a single unit.

Steps

  1. 1

    Path 1: In the dashboard

    Navigate to Documents in the dashboard. Click "Upload document", choose a file, optionally override the title and confirm.

    Outcome: The document appears in the list with status Uploaded, shifts to Indexed after a few seconds, and is then searchable with status Ready.

  2. 2

    Path 2: From the command line

    bash
    # Single file
knowmind upload ./onboarding-handbook.md --title "Onboarding Handbook v3"

# All Markdown files from a directory (idempotent, content-hash based)
knowmind sync ./docs --manifest ./.knowmind.manifest

    knowmind sync computes a content hash per file and transfers only what changed since the last run. Choose the manifest path freely; keep it inside the project but exclude it from version control.

    Outcome: For each uploaded document the command prints a success message with memory ID and the number of generated sections.

  3. 3

    Path 3: Via the API

    For your own applications the upload goes to POST /api/v1/ingest/document:

    bash
    curl -X POST https://knowmind.de/api/v1/ingest/document \
  -H "Authorization: Bearer kmt_…" \
  -H "Content-Type: application/json" \
  -d '{
    "title": "QM Handbook",
    "content": "<complete Markdown / text content>",
    "source": "internal-handbook-v3"
  }'

    For large documents the chunking runs asynchronously — the API returns a document_id right away and the status appears in the documents list.

  4. 4

    Upload a PDF

    The dashboard accepts PDFs directly. knowmind extracts the text layer; for scanned PDFs without OCR this does not work reliably. If the result is incomplete, run the PDF through OCR (Adobe Acrobat, Tesseract) first and upload the result as Markdown or text.

Verify the result

  • The dashboard Documents list shows status Ready.
  • The Memories statistics tile rose by the number of sections created.
  • A search for a phrase from the document returns at least one section with a source reference back to the document.

Troubleshooting

Error messageCauseResolution
Upload fails with 'file too large'Document larger than 10 MB.Split the document into two or more parts and upload separately.
Status stuck on 'Indexed'Embedding step still running in the background, or the embedding provider is unreachable.With BYOK setup, check the stored key (Dashboard → BYOK Keys). With the knowmind default embedding, wait a few minutes and reload.
PDF returns empty contentScanned PDF without an OCR text layer.Run OCR with a separate tool and upload the result as Markdown or text.
knowmind sync: 'permission denied'Manifest file is not writable.Check the manifest file permissions, choose a different path, or delete the manifest so it is recreated.

Related