Creating Entries from Data
Obelisk lets you create entries from unstructured text or uploaded documents. Instead of manually filling out forms, you can paste text, upload receipts, or provide any relevant information—Obelisk extracts the structured data for you.
How It Works
When you provide data to create entries, Obelisk follows a multi-step process:
- Validation: Obelisk checks if your data is relevant for the entry type and identifies how many distinct records can be extracted.
- Field Extraction: For each valid record, Obelisk extracts values for your entry type's fields, ensuring required fields are present.
- Record Creation: Valid records are created as entries in your organization. Invalid records are skipped with warnings explaining what's missing.
Example:
If your entry type is "Receipt" with fields vendor_name (required) and amount (required), you could paste:
Coffee Shop - $12.50 Gas Station - $45.00 Restaurant - $67.89
Obelisk would create three Receipt entries, extracting the vendor name and amount from each line.
What You Can Provide
You can create entries from two types of data:
Any text containing relevant information. This can be lists, paragraphs, bullet points, or any format that includes the data you need extracted. Obelisk will identify distinct records and extract field values.
Images or PDFs containing structured data like receipts, invoices, or forms. Obelisk extracts text from documents and then processes it like unstructured text.
Important Constraints
To ensure quality, several constraints apply when creating entries from data:
Required Fields Must Be Present
Every record must include all required fields for your entry type. If a record is missing required information, it will be skipped with a warning. You'll see which records were created successfully and which were skipped.
Example:
For a Receipt entry type requiring vendor_name and amount:
Receipt 1: Coffee Shop - $12.50 ✓ Valid Receipt 2: Unknown vendor - no amount ✗ Skipped (missing amount) Receipt 3: Target - $45.99 ✓ Valid
Result: 2 entries created, 1 skipped with warning
Data Must Be Relevant
Obelisk validates that your data matches the entry type before processing. If the data appears unrelated to the fields in your entry type, the process will fail with a relevance error.
Maximum 100 Records Per Submission
You can create up to 100 entries in a single submission. If Obelisk detects more than 100 distinct records in your data, the process will abort without creating any entries. Break your data into smaller batches if you need to create more.
Required Relationships Must Be Provided
If your entry type has required relationships to other entry types (e.g., every Receipt must belong to a Company), you must manually provide these relationships before creating entries from data. Obelisk does not automatically create or infer parent entries.
Example:
If Receipt requires a relationship to Company, you must select the Company first when creating receipts from data. The form will prompt you to select the required relationship before proceeding.
Error if not provided: "Missing required relationships: Company"
Best Practices
To get the best results when creating entries from data:
- Be specific: Include clear labels or context for your data. "Coffee Shop - $12.50" is better than just "$12.50".
- Format consistently: Use similar formatting for each record to help Obelisk identify patterns and extract data accurately.
- Provide complete information: Ensure each record includes all required fields. Incomplete records will be skipped.
- Review drafts: Always review draft entries before publishing to verify Obelisk extracted data correctly.
- Use appropriate entry types: Create entry types with fields that match the data you typically work with for best extraction results.
Troubleshooting
Your data doesn't match the fields defined in this entry type. Double-check you selected the correct entry type, or restructure your data to include the relevant fields.
Split your data into multiple submissions of 100 records or fewer.
Select the required parent entry before submitting. The form will show which relationships are required.
Check which required fields are missing from skipped records. Successfully created entries are saved as drafts regardless of warnings.