Platform Guide
Turn unstructured documents into validated, structured data with per-cell provenance. Sources bring documents in, the Field Registry learns the fields across them, a Spec composes a rail of stages over one schema, and each pipeline run of that Spec produces a Data Product you can review, deliver, and decide on with Apps.
Supported Formats
25+ file types
Guide
19 groups · 102 pages
Document Ontology
529 types
Overview
Core concepts, platform flow, and getting started guide for Talonic: ingest documents, structure them with Specs and Pipelines, and review the results.
AI Agent
Embedded AI assistant accessible via Cmd+I from any page. Inspects schemas, searches documents, analyzes extraction quality, explores cases, and builds schemas through natural language.
Sources & Documents
Upload on the Documents page via drag-and-drop, API, or connectors. 25+ formats with automatic OCR, classification pills, and in-zone upload tags.
Field Intelligence
The Field Registry with Core, Proven and Candidate maturity levels, semantic clustering, field resolution, and AI-synthesized master extraction instructions.
Schemas & Templates
AI-generated schemas per document type and user-defined templates with field matching, reference tables, and versioning.
Extraction Jobs
The quick path: a Job runs four phases (Resolve, Planner, Validation, Re-read), filling the grid progressively with per-cell provenance and confidence.
Pipelines
The Spec and the engine: compose a rail over one schema, run it per document through extraction, resolution and validation, then review in Review Mode and assemble.
Linking & Cases
Document linking discovers relationships through shared entities. Related documents grouped into cases with evidence chains.
Batch Inference
Batch processing mode defers Claude extraction to the provider batch API at 50% cost with 48-hour delivery. Monitor batch progress and results.
Smart Matching
Upload reference datasets and configure weighted field-to-field matching strategies. Run matches asynchronously and review top-5 candidates per document.
Data Products
Dataset templates for reusable output specifications and assemblies for combining multi-source documents into structured datasets.
Validation & Benchmarks
Automated validation checks, Ground Truth benchmarking, Review Gates, Field Review on Products, and Admin field deprioritization rules for the Review Queue.
Outputs
Push structured data to webhooks, REST APIs, SFTP, email, or cloud storage with field mappings, triggers, delivery configuration, and telemetry.
Apps
Turn governed data into decisions: create apps by describing them, read the verdict matrix, audit any run in the Run Inspector, and close the loop through the Human Review inbox and review teams.
Workspace Settings
Opened from the user menu: shared dialects for output formatting, reference primitives for lookup tables, change review governance, and the workspace AI policy.
Enterprise Archive
Retention policies with review-gated disposition, fail-closed legal holds, archive mode with tombstones, and a tamper-evident hash-chained audit trail.
Search & Filtering
Global omnisearch (Cmd+K) and the unified Documents facet bar: status, type, source, tag, batch and key:value filters, multi-select triage tags, saved presets.
API & Webhooks
API key management, public REST API overview, and webhook configuration for event-driven integrations.
Team & Admin
Four roles (Viewer, Member, Senior Member, Admin), team management from the user menu, credit usage tracking, the admin panel, and keyboard shortcuts.