Skip to content
Asclepius

First Steps

After installation, the first-launch setup wizard creates your admin account and first patient and logs you in. After that, three things matter before you can ingest documents.

1. Configure LLM & OCR

Section titled “1. Configure LLM & OCR”

Open Settings → Document Analysis → Providers and add:

  • Credentials, one per physical endpoint (Ollama server, Claude / OpenAI account). Each credential carries its URL, API key, concurrency cap, and retry policy.
  • LLM provider, pointing at a credential and a model (e.g. qwen2.5, claude-haiku-4-5-20251001).
  • OCR provider(s), Tesseract (built-in, no credential), LLM Vision (vision model), Google Cloud Vision, or Tesseract Remote.
  • Vision-LLM provider (optional), for the single-step vision flow.

The Priority sub-tab reorders providers for fallback.

2. Ingest a document

Section titled “2. Ingest a document”

Upload via the Documents page (drag-and-drop or file picker) or drop files into vault/inbox/. The watcher picks them up within a few seconds (configurable via pipeline.poll_interval_seconds). Supported formats: PDF, PNG / JPG / TIFF, DICOM.

3. Watch it process

Section titled “3. Watch it process”

The Dashboard’s Pipeline card shows the live job: filename, kind (Upload or Reprocess), flow type (OCR + LLM or Vision-LLM), a connected stepper across the planned stages, a live elapsed-time clock, and a shimmering page-progress bar during OCR. An “Up next” rail underneath lists what’s queued behind the running job. Each document moves through OCR → LLM extraction → organizing in the default flow; the Vision-LLM flow shortens that to Vision → LLM extraction → organizing.

The document detail page carries a per-doc Pipeline stages card that shows every upload + reprocess this document has been through, grouped by run, with durations, error messages, and outcome pills — useful when comparing reprocesses or debugging a failure.

For multi-page PDFs you may also see chunked extraction (>1 page or

8k chars), page classification, and section extraction surfaced as their own stages.

Next steps

Section titled “Next steps”