Extract

okra extract uploads, processes, and extracts structured data from a PDF in one step. The source can be a local file path, a URL, or the document ID of a PDF you already uploaded.

# From a local file
okra extract ./report.pdf

# From a URL
okra extract https://example.com/report.pdf

# From an existing document
okra extract doc-abc123

Structured extraction with a schema

Pass a JSON Schema to get structured JSON back. The schema can be a file path or inline JSON.

okra extract ./invoice.pdf \
  --schema ./invoice.schema.json \
  --prompt "Extract vendor, total, currency, and due date"

When stdout is not a TTY (or with --json), the result comes back in the standard envelope:

{
  "ok": true,
  "command": "extract",
  "result": { "data": {} },
  "cost": { "usd": null },
  "citations": [],
  "next_actions": []
}

Per-field citations (`--cite`)

Add --cite to attach per-field source citations — the page and bounding box each extracted value came from. Grounding is opt-in: without --cite you get values only.

okra extract ./invoice.pdf --schema ./invoice.schema.json --cite

Citations are returned in the Anthropic-shaped citations array (type: page_location, with the source page, bbox, and the schema field each value maps to), alongside the extracted data.

Fire-and-forget (`--no-wait`)

By default okra extract waits for processing to finish. Use --no-wait to queue the work and return immediately; follow up with okra jobs wait.

okra extract ./report.pdf --no-wait
okra jobs wait <docId>

Flags

Flag	Description
`--schema <file>`	JSON Schema file or inline JSON for structured extraction
`--prompt <query>`	Extraction prompt (default: “Extract all data according to the schema”)
`--cite`	Return per-field source citations (page + bbox) for each extracted value
`--no-wait`	Fire-and-forget — queue the job and don’t wait for processing

Global flags (--json, --quiet, --output) apply as well — see the CLI Reference.

To browse a document’s text, tables, and structure after extraction, use the grounded context commands (okra context structure, okra context tables, okra context get) or read raw markdown with okra read <docId>.

Getting Started

Documents

Jobs

Collections

Reference

Structured extraction with a schema

Per-field citations (`--cite`)

Fire-and-forget (`--no-wait`)

Flags

​Structured extraction with a schema

​Per-field citations (--cite)

​Fire-and-forget (--no-wait)

​Flags

Structured extraction with a schema

Per-field citations (`--cite`)

Fire-and-forget (`--no-wait`)

Flags