MotherDuck connects local DuckDB analysis to a shared cloud warehouse

MotherDuck is a cloud data warehouse built on DuckDB. It adds managed storage, sharing, administration and cloud compute to a SQL engine many developers already use locally. Its architecture can coordinate local and cloud query execution, while newer features support scheduled Python pipelines and reusable guidance for AI agents. The practical appeal is a path from a useful local analysis to a shared, maintained data product.

01 / ProductDuckDB with cloud storage and shared access

The architecture guide³ describes a service that runs DuckDB SQL against managed data and supported external sources. Developers can connect through DuckDB clients, use the web interface or select other supported access routes. The service supplies identity, permissions, catalog management and operational capabilities around the query engine.

Dual Execution can place query stages in local DuckDB, MotherDuck’s cloud engine or both. For example, a local file can be joined with a cloud table without the developer first building a separate upload workflow for every exploratory query. The query and data placement still matter: local access is not a promise that all information remains on the laptop.

MotherDuck calls its compute instances Ducklings. Its architecture isolates compute by user or agent, and larger plans provide different instance sizes and read-scaling options. That is a different operating model from configuring one shared cluster for every analyst. The team still needs to understand which identities and workloads consume resources.

02 / AudienceWho should consider MotherDuck

MotherDuck is relevant to SQL-oriented teams that want managed analytics without starting with a large infrastructure project. It can suit internal reporting, product analytics and customer-facing data applications. A team already using DuckDB for files or notebooks may find the connection between local work and shared cloud data especially useful.

The platform is also relevant when agents need access to a maintained analytical dataset. Its MCP server and Guides can provide context around tables and business definitions. That capability becomes more useful when the underlying data is already well modeled; connecting an assistant does not resolve ambiguous joins or inconsistent customer identifiers.

The Hex blueprint offers a complementary analysis and application layer. The Snowflake blueprint is a useful warehouse comparison when broader platform requirements, organizational scale or an existing ecosystem drive the choice. Compare workload behavior and operating needs rather than treating warehouse size as a measure of quality.

03 / WorkflowA proposed workflow for customer-facing usage analytics

Imagine a software company showing each customer how its teams use the product. This is a proposed workflow, not a completed implementation. Define the event grain first: one row per action, session or daily aggregate. Identify the customer key, event time and rules for excluding internal test accounts. Keep these definitions visible before building charts.

Load a representative event dataset and create a small set of aggregate tables. Use local DuckDB for exploratory work where convenient, then move the accepted transformation into a maintained cloud workflow. Check that repeated imports do not duplicate events and that late-arriving records update the intended reporting period. Preserve the source event identifier for reconciliation.

Consider Flights⁶ for scheduled ingestion or Python transformations. A Flight is a Python program that runs in an isolated runtime and connects to MotherDuck through a Duckling for SQL work. The architecture documentation marks Flights as preview. Review generated code, pin relevant dependencies and define how a failed run resumes without reapplying already completed work.

Use sharing controls⁴ to design access before exposing the analytics. MotherDuck supports sharing with the organization, specific users or roles, and public access within the same region. A public share is not appropriate merely because an embedded dashboard is convenient. Test two customer identities and prove that each sees only the intended tables and records through the chosen application design.

Write Guides⁵ for the definitions an agent cannot infer from column names. A guide might explain what counts as an active team, which events are internal and how to join account mappings. Guides are preview Markdown context that connected agents can discover. Keep organization-wide definitions separate from customer-specific context, especially when several end users share a service-account connection.

Expose a small set of reviewed charts or an analytical interface, with freshness and reporting-period labels. Compare the displayed totals with source records and test an empty customer, a large customer and a customer whose identifier changed. The first release should prove that the data refreshes, the definitions are stable and customer access is correctly scoped before adding open-ended AI questions.

04 / PricingPricing combines an organization fee with regional usage

Pulse and Standard do not meter the same thing. Pulse charges compute units consumed by queries, with a minimum unit duration. Non-Pulse instances use wall-clock time and a configurable cooldown period after work completes. A query that takes a few seconds can therefore have a different billing pattern depending on the chosen instance type.

Flights add Python runtime charges alongside the instance work used for SQL. The detailed documentation makes Flights available across plans, with Lite scheduling and limits depending on whether a payment card is on file. This is more specific than the plan summary alone. Check the intended scheduling and run-duration requirements before budgeting a recurring import.

Storage, AI functions and read-scaling replicas are further categories. Internal users and application end users are also different plan concepts. For customer analytics, estimate refresh work separately from interactive reads, then measure usage by the application’s actual identities and traffic. A platform fee does not make each customer dashboard query free.

05 / DistinctionsWhat stands out in the analytics workflow

The local-and-cloud relationship is a useful differentiator. Analysts can retain familiar DuckDB tools while sharing accepted data and results through the cloud service. That can shorten the distance between an exploratory query and a maintained dataset, provided the final transformation and refresh process are recorded rather than left only on one laptop.

Guides address a concrete weakness in natural-language analytics: business knowledge often lives outside the schema. A reusable explanation of account identity or event exclusions can help multiple agent sessions begin with the same context. The guide still needs an owner and review when the tables or business rules change.

Flights extend the platform beyond SQL execution into scheduled Python work. This can consolidate a small ingestion or enrichment pipeline near the data. It also introduces a second execution plane, with its own runtime, dependencies and billing. The benefit is most clear when that program has a defined purpose and observable completion criteria.

06 / QuestionsQuestions to settle before sharing production data

Which region will the organization use? MotherDuck organizations are scoped to one cloud region selected at creation, and shares are region-scoped. The current architecture lists six AWS regions. Region is therefore an early design choice, affecting access arrangements, pricing and where the service operates.

Where do AI functions process data? The architecture documentation explicitly says that AI functions for the Asia Pacific regions currently use a US-based endpoint. An organization’s warehouse region is not sufficient evidence for every AI processing route. Check the selected function and current regional documentation before using it on location-restricted data.

What does a private Guide mean under a shared service account? The Guides documentation says user-private context belongs to the connection identity. If several end users share that identity, the guide is shared context for those users. Topic names organize information; they are not an additional authorization boundary. Design customer-specific guidance with the actual connection model in mind.

07 / DecisionChoose MotherDuck for a maintained SQL data product

MotherDuck is a strong candidate when DuckDB-style analysis needs managed storage, collaboration and a route into an application. Start with one dataset and make the refresh, definitions and access model explicit. Use Flights or Guides when they solve a concrete pipeline or context problem, with their current preview status understood.

For customer-facing analytics, the useful outcome is a reliable, correctly scoped view of product usage with known freshness and measurable cost.