# Zingg Entity Resolution

> Warehouse-native entity resolution that matches customer, supplier and product records with ML in your own data platform (Databricks, Fabric, AWS, GCP or Snowflake); open source, with paid editions for persistent IDs.

Source: https://toolsfordata.com/tools/zingg/ · Updated 2026-11-19

[Website](https://www.zingg.ai/) · [Docs](https://docs.zingg.ai/zingg) · [Pricing](https://www.zingg.ai/product/pricing) · [GitHub](https://github.com/zinggAI/zingg/) (1.3k stars)

Best for: data engineering teams building customer 360 views, MDM or composable CDPs who want ML record matching inside their existing cloud data platform.

## Overview

- Core Strength: Warehouse-native ML entity resolution in Databricks, Fabric, AWS, GCP or Snowflake
- Matching: ML model trained from 40–50 labeled pairs; rules in paid editions
- Golden Records: None yet: stewardship and golden records announced
- Platform Fit: Databricks, Fabric, AWS, GCP; Snowflake Native App in paid editions
- Deployment Model: Self-hosted in your own cloud account
- Pricing: Free (AGPL-3.0); paid editions priced per deployment
- Notable Caveat: AGPL-3.0; persistent IDs and incremental runs need a paid edition

## What it's used for

Zingg is used for warehouse-native entity resolution: it matches records about the same customer, household, supplier or product with ML inside your own data platform (Databricks, Fabric, AWS, GCP or Snowflake), for a customer 360 view or the matching step of master data management.

## Pricing

Zingg Community is free under AGPL-3.0. Two paid editions are priced per deployment: Enterprise Lite adds a persistent ZINGG_ID, incremental runs and the Snowflake Native App; Enterprise adds explainability, model comparison, nickname dictionaries and standardization. Stewardship, golden records and real-time matching are announced as coming soon.

## Deployment

Zingg runs as Spark jobs inside your own account: Databricks (including Unity Catalog), Microsoft Fabric, AWS Glue or EMR, GCP Dataproc, or any Spark cluster. The paid editions also run natively in Snowflake as a Snowflake Native App. Data never leaves your platform.

## Editions

| Edition | Price | What it includes |
|---|---|---|
| Community | Free (AGPL-3.0) | ML matching trained on your labels, batch runs on Spark in Databricks, Fabric, AWS or GCP, connectors to Snowflake, Redshift and BigQuery, and a Z_CLUSTER ID rebuilt on every run. |
| Enterprise Lite | Per deployment | Everything in Community, plus a persistent ZINGG_ID, incremental runs on new and changed records, deterministic rules in the same flow as the model, pass-through fields, the Snowflake Native App and enterprise support. |
| Enterprise | Per deployment | Everything in Enterprise Lite, plus ID reassignment after retraining, nickname dictionaries, standardization, explainability and match statistics, model comparison and a knowledge graph. Real-time resolution, streaming, stewardship and golden records are listed as coming soon. |

## Review

Strengths:
- Warehouse-native: matching runs inside Databricks, Fabric, AWS or GCP, or natively in Snowflake in the paid editions, and IDs are written back to your tables.
- ML matching trained from 40 to 50 labeled pairs, with learned blocking that scales to millions of records.
- Free Community edition under AGPL-3.0, and the paid editions use the same configuration and Python API.

Caveats:
- AGPL-3.0 can block companies that ship it inside their own products.
- The free Z_CLUSTER ID changes between runs; a persistent ZINGG_ID and incremental runs need a paid edition.
- Entity resolution today: stewardship and golden records are announced, not shipped.

Recommendation: Pay for Enterprise Lite when IDs must stay stable between runs. For probabilistic matching without labeling, consider Splink; for real-time resolution inside an application, Senzing; for a full MDM hub with stewardship, Tamr, Reltio or Informatica.

## Features

### Matching

- ML Matching: Yes — Community: "Probabilistic AI matching trained on your own labels"; blocking is learned too. ([source](https://www.zingg.ai/product/pricing))
- Rule-Based Matching: Yes — Enterprise Lite and up: deterministic exact-match conditions run next to the ML model in one flow. ([source](https://docs.zingg.ai/zingg/running-zingg/set-deterministic-matching-conditions.md))
- Real-Time Matching: No — Real-time resolution and streaming are "Coming soon" in Enterprise. ([source](https://www.zingg.ai/product/pricing))
- Incremental Matching: Yes — Enterprise Lite and up: the incremental flow handles "new, changed, or deleted records" without a full rematch. ([source](https://docs.zingg.ai/zingg/zingg-concepts/zingg-entity-resolution-platform/incremental-run.md))
- Persistent Entity ID: Yes — Enterprise Lite and up: ZINGG_ID, "a persistent globally unique identifier". The free Z_CLUSTER is rebuilt every run. ([source](https://docs.zingg.ai/zingg/zingg-concepts/z-cluster-and-zingg-id.md))
- Match Explainability: Yes — Enterprise: "Explainability and match statistics": which fields and scores drove each match. ([source](https://docs.zingg.ai/zingg/community-and-enterprise-editions/community-vs-enterprise.md))

### Golden Records

- Survivorship Rules: No — "Golden records" are "Coming soon"; output is a cluster ID, and building the record is left to you. ([source](https://www.zingg.ai/product/pricing))
- Survivorship on Read: No — Zingg outputs a cluster ID; it doesn't store or serve golden records. ([source](https://www.zingg.ai/product/pricing))
- Stewardship Review: No — "Data stewardship" is "Coming soon" in Enterprise. ([source](https://www.zingg.ai/product/pricing))
- Record History: No — No record version or merge history is documented. ([source](https://docs.zingg.ai/zingg/llms.txt))
- Hierarchy Management: No — No master-record hierarchies; the knowledge graph recipe analyzes relationships between resolved entities. ([source](https://docs.zingg.ai/zingg/llms.txt))

### Domains & Modeling

- Multi-Domain MDM: No — Each run builds a model for one dataset; no stored, managed domains are documented. ([source](https://docs.zingg.ai/zingg))
- Configurable Data Model: No — You pick the fields to match for any entity, but there is no stored master data model of entities and relationships. ([source](https://www.zingg.ai/product/pricing))
- Product Information Management: No — No PIM features; product data is matched, not managed. ([source](https://docs.zingg.ai/zingg/llms.txt))
- Reference Data Management: No — Mapping files standardize output values; no managed code lists. ([source](https://docs.zingg.ai/zingg/running-zingg/standardize-fields-and-results.md))
- Master Data Authoring: No — No entry UI; Zingg runs batch and incremental jobs. ([source](https://docs.zingg.ai/zingg/llms.txt))

### Data Quality

- Validation at Entry: No — Zingg runs batch and incremental jobs; there is no entry UI or API. ([source](https://docs.zingg.ai/zingg/llms.txt))
- Standardization: Yes — Enterprise: "Standardise fields before matching", with nickname and alias dictionaries. ([source](https://docs.zingg.ai/zingg/community-and-enterprise-editions/community-vs-enterprise.md))
- Third-Party Enrichment: No — No enrichment connectors are documented. ([source](https://docs.zingg.ai/zingg/llms.txt))

### Data Distribution

- REST API: No — The interface is a Python API, not a REST API. ([source](https://docs.zingg.ai/zingg/llms.txt))
- Change Events: No — No event publishing is documented. ([source](https://docs.zingg.ai/zingg/llms.txt))
- Warehouse Write-Back: Yes — Zingg "writes resolved IDs back to your tables" in Databricks, Snowflake, Fabric or GCP, in every edition. ([source](https://www.zingg.ai/product/pricing))
- Data Syndication: No — No publishing to external channels. ([source](https://docs.zingg.ai/zingg/llms.txt))

### AI & Agents

- MCP Server: No — No MCP server; the closest is an identity RAG recipe with LangChain. ([source](https://docs.zingg.ai/zingg/llms.txt))
- AI Stewardship Assistance: No — No AI assistant for steward decisions is documented. ([source](https://docs.zingg.ai/zingg/llms.txt))

### Platform Native

- In-Place Matching: Yes — Runs as Spark jobs in your own Databricks, Fabric, AWS or GCP account, or as a Snowflake Native App (Enterprise Lite and up); IDs are written back to your tables. ([source](https://docs.zingg.ai/zingg/community-and-enterprise-editions/community-vs-enterprise.md))
- Snowflake Native: Yes — Enterprise Lite and up: "Snowflake Native App on Snowflake Marketplace", no Spark cluster needed. ([source](https://docs.zingg.ai/zingg/community-and-enterprise-editions/community-vs-enterprise.md))
- Databricks Native: Yes — Community and up run as jobs in your Databricks workspace, with Unity Catalog support since 0.6.0. ([source](https://docs.zingg.ai/zingg/community-and-enterprise-editions/community-vs-enterprise.md))
- Microsoft Fabric Native: Yes — Community and up run on Fabric Spark notebooks in your workspace. ([source](https://docs.zingg.ai/zingg/community-and-enterprise-editions/community-vs-enterprise.md))

### Deployment

- SaaS Deployment: No — "Zingg runs inside your Databricks, Snowflake, Microsoft Fabric, Google Cloud or AWS account"; there is no hosted service. ([source](https://www.zingg.ai/product/pricing))
- Self-Hosted Deployment: Yes — Runs on Spark in your own account (Databricks, Fabric, AWS Glue or EMR, GCP Dataproc) or Docker. ([source](https://www.zingg.ai/product/pricing))

### Open Source

- Open Source: Yes — "Zingg is licensed under AGPL v3.0". ([source](https://github.com/zinggAI/zingg))

## Integrations

24 integrations, each sourced from the vendor docs.

- AI Agents: LangChain
- Data Governance: Databricks Unity Catalog
- Data Platform: Databricks, Microsoft Fabric, AWS Glue, Amazon EMR, Google Cloud Dataproc, Apache Spark
- Data Storage: Amazon S3, Azure Blob Storage, Google Cloud Storage
- Data Warehouse: Snowflake, Google BigQuery, Amazon Redshift, Exasol
- Databases: PostgreSQL, MySQL, SQL Server, IBM Db2, MariaDB, Oracle Database, Apache Cassandra, MongoDB, Neo4j

## Alternatives

- [Splink](https://toolsfordata.com/tools/splink/): The closest open-source match: explainable probabilistic matching in Python, on DuckDB or Spark.
- [Senzing](https://toolsfordata.com/tools/senzing/): Consider Senzing when records must resolve one by one in real time inside your own application; its rules are built in, so there's nothing to label.
- [Tamr](https://toolsfordata.com/tools/tamr/): Consider Tamr when matching should feed golden records and curator review: ML pair scoring like Zingg, but records are loaded into Tamr.
- [Reltio by SAP](https://toolsfordata.com/tools/reltio/): Consider Reltio when operational systems need real-time golden records from a SaaS hub, and copying records into its tenant is acceptable.
- [Informatica IDMC by Salesforce](https://toolsfordata.com/tools/informatica/): Consider Informatica when matching is one part of a full MDM suite with hierarchies, stewardship and 360 apps, run in its cloud.
