Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.getclaro.ai/llms.txt

Use this file to discover all available pages before exploring further.

Claro is a platform for keeping product and supplier catalogs clean, structured, and decision-ready — at the scale and frequency that real catalogs actually change. Most catalog tooling assumes data arrives once, gets cleaned once, and stays clean. In practice, catalogs are continuously edited by suppliers, partners, and internal teams; attributes drift; new SKUs arrive without specs; taxonomies evolve; duplicates accumulate. Claro is built around that continuous reality. The platform is organized around two complementary surfaces:
  • A persistent Catalogue — your products, suppliers, SKUs, and the relationships between them, with full attribute schemas, history, and provenance.
  • Operations that run on the Catalogue — enrichment, validation, normalization, deduplication, taxonomy generation, monitoring, and sync — each producing auditable changes with confidence scores attached.
Research Agents (lists, document parsing, scraping, spreadsheet analysis) remain available for one-off datasets and exploration, but the center of gravity of the platform is now the Catalogue and the operations that act on it.

​
Core concepts

​
Catalogue

A Catalogue is a typed object — Products, Suppliers, Line Items, Variants, etc. Each object has:
  • Attributes — the schema (typed fields: text, number, image, date, enum, reference).
  • Records — the actual rows.
  • Data Sources — the upstream feeds, files, APIs, or scrapes that populate it.
  • Operations — the running and scheduled transformations against it.
  • Config — access, retention, write-back rules.
Multiple catalogues can exist in a workspace and can reference each other (e.g. Product.supplier → Supplier).

​
Operations

Operations are the verbs of the platform. They read records, optionally call models or external systems, and write back changes. Every operation produces a run with inputs, outputs, confidence per change, and a reversible diff. Operations can be triggered manually, scheduled, or chained into pipelines.

​
Confidence and review

Every model-driven write is scored. Each operation has a configurable confidence threshold:
  • Above threshold → auto-apply
  • Below threshold → queue for human review in the Notifications surface
Thresholds, reviewers, and auto-apply behavior are set per operation and per attribute.

​
Provenance

Every value in the Catalogue carries the source of the last write — which operation, which run, which underlying data source, which model, which reviewer. Roll-back is per-field.

​
Taxonomy

Taxonomies are first-class objects, not free-text labels. They are hierarchical, versioned, and can be generated from your data, imported from a standard (GS1, Google, internal), or hand-edited. Taxonomy assignments are themselves tracked operations with confidence scores.

​
Similarity & Duplicate graph

Claro maintains a similarity graph across records using a combination of embedding similarity and rule-based matching. Duplicate detection, supplier match, variant grouping, and entity resolution all read from this graph.

​
Platform layout

The product is organized into three groups in the sidebar: 
CORE          — the persistent catalog and its global views
MODULES       — the operational stages: Onboard, Analyse, Monitor, Distribute
RESEARCH AGENTS — one-off list-building and extraction tasks
See Platform Overview for a tour of every surface.

​
What changed from the previous version

If you used Claro before, the most important changes:
  • Catalogues are now the center of the platform. What used to be one-off datasets generated by Research Agents now sit alongside persistent catalogues with attributes, records, data sources, operations, and config.
  • Operations are first-class. Every transformation — enrichment, validation, normalization, deduplication, taxonomy assignment, sync — is an auditable run with confidence scores and a reversible diff.
  • The sidebar is restructured into Core (persistent state), Modules (operational stages: Onboard, Analyse, Monitor, Distribute), and Research Agents (ad-hoc tasks).
  • Continuous monitoring is built in. Price, competitor, and validation operations can run on a schedule and alert into Notifications.
  • Confidence and review are explicit. Every operation has thresholds; uncertain changes queue for human approval rather than failing silently or shipping unreviewed.
Ready to set up a catalogue? Continue to Quick Start Guide.