• Document Field Extraction — Upload a PDF or document. The AI reads the content, identifies every form field, and extracts labels, types, values, and groupings into a structured format — ready for automated processing.
• Auto-Fill from Company Knowledge — The extracted fields are matched against your stored company data — addresses, tax IDs, certifications, reference projects, and more. Each field is filled automatically with confidence scores so you know what to review.
• Side-by-Side Comparison — See the original extraction and AI-filled results side by side, making it easy to review, correct, and approve before submission.
• Configurable Extraction — Customize how fields are extracted and filled through the Operaide UI — no code changes needed. Adjust extraction rules, field handling, and output formatting to match your specific forms.

• File Injection into LLM Context — aktorBase64ToFile converts the uploaded base64 data URL into a proper File object with auto-detected MIME type and filename. aktorInjectFile then converts it into a FilePart message compatible with the Vercel AI SDK, injecting the file directly into the LLM's message history.
• Schema-Enforced Generation — aktorAIGenerateObject calls the LLM with a strict Zod schema (DocumentExtractionSchema). The framework validates the LLM output against the schema — guaranteeing type-safe results with proper field types, confidence scores, and groupings. No post-processing or manual parsing needed.
• Two-Phase Pipeline — Phase one extracts raw form structure from the document. Phase two retrieves relevant company data from AgentDB, injects it alongside the extraction result into a fill prompt via aktorCompletePrompt, and calls aktorAIGenerateObject again to populate ai_filled_value fields. Both phases use the same schema for consistent typing.
• Rich Field Type System — The extraction schema supports 14 field types (text, number, currency, date, checkbox, radio, select, table, and more), hierarchical field groupings, optional enums, and confidence scoring from 0 to 1 — providing a comprehensive representation of any form structure.
• UI Schema Tags — The [file-upload] tag in the input schema renders a file picker in the Operaide UI. The [textarea] tag on output fields renders multi-line display areas for the raw and filled JSON — enabling side-by-side review directly in the platform.

The Document Fill app registers two Reaktors. The extract-and-fill Reaktor is the full pipeline: the uploaded base64 file enters aktorBase64ToFile, gets injected into the message history via aktorInjectFile, and flows into aktorAIGenerateObject with a system prompt and the DocumentExtractionSchema for field extraction. Company knowledge is retrieved from AgentDB and injected alongside the extraction result into a fill prompt via aktorCompletePrompt, then passed to a second aktorAIGenerateObject call that populates ai_filled_value fields. Finally, aktorJoinObjects combines the raw and filled outputs into a single response. The doc-parsing Reaktor is a simpler variant that performs only the extraction phase — useful for cases where auto-fill isn't needed.