Settings¶
OntExtract provides configurable settings to customize application behavior. Access settings via Settings in the navigation menu (admin users only).
Settings Categories¶
LLM Integration¶
Controls LLM-assisted features like template enhancement.

| Setting | Type | Default | Description |
|---|---|---|---|
| Enable LLM Enhancement | Boolean | On | Enable LLM template enhancement for experiment descriptions |
| Max Tokens | Integer | 200 | Maximum response length for template enhancement (50-1000) |
API Key Required
LLM features require the ANTHROPIC_API_KEY environment variable to be set.
NLP Tools¶
Configure the transformer-based definition extraction tool.
| Setting | Type | Default | Description |
|---|---|---|---|
| Confidence Threshold | Float | 0.70 | Minimum confidence for definition extraction (0.5-0.95). Higher = fewer but more accurate results |
Prompt Templates¶
Manage Jinja2 templates for experiment descriptions. Templates can be viewed and edited, with optional LLM enhancement support.
| Column | Description |
|---|---|
| Template | Template key identifier |
| Category | Template category (e.g., experiment descriptions) |
| Variables | Required template variables |
| LLM Enhancement | Whether LLM can enhance template output |
| Actions | View or edit the template |
Processing¶
Configure how documents are processed during text cleanup operations.

| Setting | Type | Default | Description |
|---|---|---|---|
| Concurrent Chunk Processing | Boolean | On | Process document chunks in parallel for faster cleanup |
| Max Concurrent Chunks | Integer | 3 | Maximum simultaneous API calls (1-10) |
How Text Cleanup Works¶
When cleaning large documents (>8,000 characters), OntExtract:
- Chunks the document by paragraph boundaries
- Processes each chunk through Claude for OCR correction, spelling fixes, and formatting improvements
- Reassembles cleaned chunks in order
Parallel Mode (default): Multiple chunks processed simultaneously. Faster overall but uses more concurrent API calls.
Sequential Mode: Chunks processed one at a time. Predictable progress, lower concurrent API usage.
Recommended Settings
The default of 3 concurrent chunks balances speed with API rate limits. Increase to 5-10 for faster processing with higher rate limits.
Provenance¶
Configure how provenance records are handled during deletions.
| Setting | Type | Default | Description |
|---|---|---|---|
| Purge on Delete | Boolean | On | Permanently delete provenance records (vs. mark as invalidated) |
| Show Deleted in Timeline | Boolean | Off | Display invalidated items in timeline views by default |
When Purge on Delete is disabled: - Provenance records are marked as "invalidated" but preserved - Complete audit trail is maintained - Deleted items can be shown in timeline views
Saving Settings¶
- Navigate to the Settings page
- Select the appropriate tab (LLM Integration, NLP Tools, Prompt Templates, Provenance, or Processing)
- Modify settings as needed
- Click Save Changes
Settings are stored per-user, allowing different users to have different preferences.
Related Guides¶
- LLM Orchestration - Using LLM-powered document analysis
- Process Documents - Text cleanup and processing
- Provenance Tracking - Understanding the audit trail