Manage files, extractors, and parsers

Files

Upload and capture the returned metadata:

parsehawk files upload document.pdf
parsehawk files list
parsehawk files get file_...

Delete a stored file when no longer needed:

parsehawk files delete file_...

An existing job retains its record, but deleting source content can prevent new work or later content retrieval. Apply a retention policy to both files and job results.

Extractors

Use a stable, lowercase name with ASCII letters, digits, hyphens, or underscores. It cannot begin with the reserved extractor_ prefix.

For source-controlled definitions, keep instructions, schema, and examples in files and synchronize idempotently:

parsehawk extractors put invoice_v1 \
  --display-name "Invoice extractor" \
  --instructions instructions.txt \
  --schema invoice.schema.json \
  --examples examples.json

Use a partial update for an intentional one-field change:

parsehawk extractors update invoice_v1 --display-name "Invoices"

Inspect and remove definitions:

parsehawk extractors list
parsehawk extractors get invoice_v1
parsehawk extractors delete invoice_v1

Version definitions in Git

Treat the files passed to extractors put as the reviewable source of truth. When the output shape changes incompatibly, create a new stable extractor name instead of silently replacing the contract used by existing consumers.

Few-shot file examples reference uploaded file_... IDs and are therefore tied to the target ParseHawk data store. Inline-text examples are easier to move between installations.

Parsers

Parser names follow the same stable lowercase convention and cannot begin with the reserved parser_ prefix. Create or synchronize a custom parser with:

parsehawk parsers put legal-document \
  --display-name "Legal document" \
  --instructions instructions.txt \
  --provider openai \
  --model gpt-5-mini

Inspect, partially update, or delete it with:

parsehawk parsers list
parsehawk parsers get legal-document
parsehawk parsers update legal-document --display-name "Legal documents"
parsehawk parsers delete legal-document

The seeded document-to-markdown parser is read-only. Parser edits affect only new jobs because every parse job stores a configuration snapshot. A parser or file referenced by a parse job cannot be deleted until the job is removed.

See customize Markdown parsing for instruction and model guidance. See operate asynchronous jobs for execution and retention behavior after resources exist.