reducto

Troubleshooting

If requests fail, run zero doctor check-connector --env-name REDUCTO_TOKEN or zero doctor check-connector --url https://platform.reducto.ai/parse --method POST.

How to Use

Reducto authenticates with a Bearer token. The base URL is https://platform.reducto.ai. Most flows upload a file (or pass a URL), get back a structured document, optionally then chunk it for RAG.

1. Parse a public PDF URL

Write to /tmp/reducto_request.json:

{
  "document_url": "https://arxiv.org/pdf/1706.03762.pdf",
  "options": {
    "ocr_mode": "agentic",
    "extraction_mode": "ocr"
  }
}

Then run:

curl -s -X POST "https://platform.reducto.ai/parse" \
  -H "Authorization: Bearer $REDUCTO_TOKEN" \
  -H "Content-Type: application/json" \
  -d @/tmp/reducto_request.json

2. Upload + parse a local file

# Step 1: upload
UPLOAD=$(curl -s -X POST "https://platform.reducto.ai/upload" \
  -H "Authorization: Bearer $REDUCTO_TOKEN" \
  -F "file=@/tmp/contract.pdf")
PRESIGNED=$(echo "$UPLOAD" | jq -r '.presigned_url')

# Step 2: parse using the presigned URL Reducto returned
cat > /tmp/reducto_request.json <<EOF
{"document_url": "$PRESIGNED"}
EOF

curl -s -X POST "https://platform.reducto.ai/parse" \
  -H "Authorization: Bearer $REDUCTO_TOKEN" \
  -H "Content-Type: application/json" \
  -d @/tmp/reducto_request.json

3. Chunk a parsed document for RAG

Write to /tmp/reducto_request.json:

{
  "document_url": "https://arxiv.org/pdf/1706.03762.pdf",
  "options": {
    "chunking": {
      "chunk_mode": "section",
      "chunk_size": 1024
    }
  }
}

Then run:

curl -s -X POST "https://platform.reducto.ai/parse" \
  -H "Authorization: Bearer $REDUCTO_TOKEN" \
  -H "Content-Type: application/json" \
  -d @/tmp/reducto_request.json

4. Structured extraction with a schema

Write to /tmp/reducto_request.json:

{
  "document_url": "https://example.com/invoice.pdf",
  "schema": {
    "type": "object",
    "properties": {
      "invoice_number": {"type": "string"},
      "total_amount": {"type": "number"},
      "line_items": {"type": "array"}
    }
  }
}

Then run:

curl -s -X POST "https://platform.reducto.ai/extract" \
  -H "Authorization: Bearer $REDUCTO_TOKEN" \
  -H "Content-Type: application/json" \
  -d @/tmp/reducto_request.json

Guidelines

1. Pick the right mode — ocr_mode: agentic is the most accurate for scans/handwriting; standard is faster and cheaper for clean PDFs.
2. Chunk mode — section (semantic), page (page-bounded), or variable (token-based) — pick based on your RAG strategy.
3. extract vs parse — extract requires a schema and returns typed JSON; parse returns the whole document as markdown + JSON.
4. Async for big docs — anything > 50 pages should use the async job endpoint; poll the job_id.
5. Multilingual works — Reducto handles CJK, Arabic, and Cyrillic out of the box.