Page Extract Python SDK

Python client for Page Extract by PageIndex: extract the data you need out of any PDF.

pip install pageindex-extract

Create an API key in the console. Page Extract and Page Search share one key, one credit balance and one set of documents.

export PAGEINDEX_EXTRACT_API_KEY=pie_live_...

The client reads PAGEINDEX_EXTRACT_API_KEY, or PAGEINDEX_SEARCH_API_KEY if you already keep the key there.

Extract

from pageindex_extract import PageExtract

client = PageExtract()

schema = {
    "type": "object",
    "properties": {
        "invoice_number": {"type": "string"},
        "total": {"type": "number", "description": "Amount due, in dollars, without the currency sign"},
        "lines": {
            "type": "array",
            "items": {
                "type": "object",
                "properties": {"item": {"type": "string"}, "amount": {"type": "number"}},
            },
        },
    },
}

job = client.extract("invoice.pdf", schema)
print(job.result)

extract uploads the file (a file uploaded before is not uploaded again), waits for the extraction and returns it. job.result is JSON that fits your schema; a field the document does not give is null.

The schema can also be a Pydantic model:

from pydantic import BaseModel

class Line(BaseModel):
    item: str
    amount: float

class Invoice(BaseModel):
    invoice_number: str
    total: float
    lines: list[Line]

job = client.extract("invoice.pdf", Invoice)
invoice = job.parse(Invoice)

Options

job = client.extract(
    "report.pdf",
    schema,
    tier="pro",            # "standard" ($0.03 a page) or "pro" ($0.05 a page)
    pages="3-9",           # only these pages, and only they are charged; also (3, 9) or 3
    cite_sources=True,     # the page each list item is on: job.sources
    system_prompt="Amounts are in thousands of dollars.",
)
print(job.usage.pages, job.usage.credits)

document can be a path, bytes, an open binary file, an uploaded document, or a document id ("doc_..."). A file is uploaded to the Page Extract collection (made on first use), where it is kept apart from documents you upload to search; pass collection="..." (a name or an id) for another, or collection=None for none. A file already in that collection is not uploaded again.

You pay only for extractions that complete. A failed extraction raises PageExtractError (its job holds the job) and costs nothing.

Jobs

Extractions run in the background. To start one and come back for it later:

job = client.extract("report.pdf", schema, wait=False)   # PENDING
job = client.jobs.wait(job)                              # COMPLETED, or raises

client.jobs.get(job.id)          # the job, with its result once COMPLETED
client.jobs.cancel(job.id)       # stop it; nothing is charged
client.jobs.delete(job.id)       # delete it and its result

for job in client.jobs.list():   # every extraction, newest first
    print(job.id, job.status)

Documents

client.documents is the Page Search client's, with the same key:

doc = client.documents.upload("report.pdf")
job = client.extract(doc, schema)

Errors

Errors raise PageExtractError with status, code, message and request_id.

Links

• Page Extract
• API docs
• Page Search Python SDK

License

MIT