Use the Web UI

The Web UI at http://127.0.0.1:5173 is a client of the same REST API used by the CLI. Resources created in one surface are immediately available in the others.

Run an extraction

1. Select the Extract workflow.
2. Upload a supported PDF, image, text, or Markdown file.
3. Choose an existing extractor, or create one with instructions and a schema.
4. Start a run with the uploaded file.
5. Inspect Extraction jobs for lifecycle state, execution metadata, and the structured result.

The seeded Receipt extractor and tests/fixtures/receipt/receipt.jpg are a known-good first path.

Create an extractor safely

• Give the extractor a stable API name that can outlive a UI label.
• Describe each schema field precisely and use nullable values for legitimate absence.
• Validate the schema before depending on it downstream.
• Add few-shot examples only when instructions and schema do not settle a recurring ambiguity.
• Select a provider and model with the capabilities required by the inputs.

The UI edits the same definition documented in core concepts.

Parse a document to Markdown

1. Select the Parse workflow.
2. Upload or choose one PDF, JPEG, or PNG.
3. Select the prebuilt Document to Markdown parser, or create a custom parser with provider, model, reasoning effort, and additional instructions.
4. Run parsing and follow the parse-job history.
5. Review rendered whole-document Markdown, switch to a one-based page view, or inspect the raw Markdown.
6. Copy the complete result or download it as a .md file.

The Document and Per page views render headings, emphasis, links, lists, code, page breaks, GitHub-style tables, and safe table HTML returned by parsing models. Raw Markdown always preserves the exact source text. Image references whose generated assets are not available are shown as labeled references instead of broken images.

The result panel also shows the provider, model, internal adapter, duration, and actionable terminal error. Active parse jobs can be canceled, and completed jobs can be deleted. The parsing token budget is server-controlled through PARSEHAWK_PARSING_MAX_TOKENS and displayed in parser settings.

For reusable parser design and model selection, see customize Markdown parsing.

Move between UI, CLI, and API

Use the stable name shown for an extractor in CLI commands:

parsehawk extract document.pdf --extractor invoice_v1 --wait

Use the stable parser name in the same way:

parsehawk parse document.pdf --parser technical-markdown --wait

Use the file_..., extractor_..., parser_..., job_..., and parse_job_... IDs shown in the UI when an API operation requires a canonical resource ID.

This page intentionally avoids step-by-step screenshots while the UI evolves. The resource names and outcomes are the stable contract; the generated REST API reference and CLI reference are authoritative for automation.