Browse, filter, and analyze all enrichment results. View detailed outputs with integrated schema navigation, grouping options, and batch operations.
The History page is the central place to review every enrichment, schema generation, playground execution, and fusion result. Records are stored in PostgreSQL and scoped to your organization. The page uses a virtualized table for smooth scrolling even with thousands of records.
Every LLM operation creates a record. The type filter lets you focus on specific kinds of results:
| Type | Source | Description |
|---|---|---|
| enrichment | Workflow Editor / Batch | Entity enrichment results with structured output matching the target schema |
| arbitration | Fusion | The merged output of a multi-model run. It has no model of its own, so the Model column names the models it merged, and the Type column says how the conflicts were settled — by deterministic rules (no LLM call, no tokens, no cost) or by an arbiter model. The Overview tab carries the full audit trail: field tally, source records, and every decision with the rule that made it. |
| schema_generation | Workflow Editor | AI-generated JSON schemas from sample data |
| schema_edit | Workflow Editor | AI-assisted schema modifications via natural language prompts |
| sample_generation | Workflow Editor | Generated sample JSON — the starting contract a schema is built from. One record covers the whole job, however many samples it produced. |
| playground | Playground | Free-form prompt executions with raw responses |
| classification | Workflow Editor / Batch | The pre-flight check asking whether the entity matches the schema's type. Cheap, non-blocking, and kept so you can see why an enrichment was warned about. |
| ambiguity_analysis | Workflow Editor | The ambiguity and identity-scoping review of a schema's properties — which names admit more than one reading, and which related items mix entity facts with pairing facts. Its findings become the advisory annotations shown in the editor. |
| db_classification | Database Sync | The link-time pass proposing each property's database key, SQL type, index and ownership. Curated in the Database Sync page's Model tab. |
Combine multiple filters to narrow down results. All filters update the table in real time:
System admins have additional filters: All organizations to view records across the entire platform, and Include deleted to see soft-deleted records.
Group records visually to spot patterns and organize results:
Flat list sorted by the selected sort option. Default view.
Group records by type (enrichment, schema generation, etc.) with collapsible sections.
Group by AI provider to compare results across different services.
Group by creation date to see daily activity patterns.
Select multiple records for batch operations using familiar keyboard shortcuts:
| Shortcut | Action |
|---|---|
Click | Select a single record and open its details |
Ctrl + Click | Toggle selection of a record without deselecting others |
Shift + Click | Select a range of records between the last click and current |
Ctrl + A | Select all visible records |
Ctrl + D | Deselect all records |
With records selected, use Batch Delete to soft-delete them. System admins can perform hard deletes for permanent removal and restore soft-deleted records.
Clicking a record opens the detail panel. Tabs appear only when the record has something to show there, so a rule-based fusion has no prompt tab and an enrichment without files has no attachments tab. The URL updates to enable deep linking (e.g., /history/uuid?tab=prompt).
The structured JSON output in a read-only Monaco editor with syntax highlighting. For enrichment records, this is the final merged result matching the schema.
View the system prompt, user prompt, and raw response for each expertise domain. Multi-expertise records show a dropdown to switch between expertise prompts.
Token usage (input, output, cached), wall-clock duration, cost breakdown, retries and validation errors. For a fusion record this is also the audit trail: the field tally, the source records, and every decision with the rule that made it.
For enrichment and fusion records: which deltas this record produced and where they stand, plus a Send to database action that re-injects the stored output into the schema's database sync.
Four more tabs open up the exact request: System Prompt, Raw Response, Wire Schema (the structured-output contract sent to the provider) and Function Tools. Plus Schema and Attachments for the inputs.
Records integrate with other parts of the application: