Running on MotherDuck?

MotherDuck hybrid queries slow or auth tokens expiring? We fix it.

MotherDuck turns DuckDB into a hosted service while keeping the local database as a first-class citizen, which means a query can scan a Parquet file on your laptop and a table in the cloud in one statement. That hybrid model is the main selling point and the main source of surprises: a plan that could have stayed local goes to the cloud, an auth token expires inside a long session, or a shared database attaches slowly from a notebook that needed sub-second startup. We trace the plan, the auth and the cloud footprint until the cost model matches the use case.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

motherduck.log

still slow

What it is

MotherDuck at a glance

MotherDuck is a managed service for DuckDB, founded in 2022 by former Firebolt and Google BigQuery leaders including Jordan Tigani. It extends the DuckDB model to a hosted database where cloud storage, team collaboration and sharing are first-class, while preserving the ability for a local DuckDB process to run queries that span local files and cloud tables.

By 2026 MotherDuck has become a common choice for small and midsize data teams that want a DuckDB-native analytics layer without giving up the laptop experience. The hybrid query planner decides which parts of a query run locally and which run in the cloud, and the pricing model charges for cloud compute rather than seats or warehouse hours. It pairs naturally with Parquet in object storage and with dbt projects that already speak DuckDB.

Type
Managed DuckDB service with hybrid local and cloud execution
First released
2023 beta, 2024 general availability
License
Proprietary cloud service, DuckDB engine is MIT
Maintained by
MotherDuck, Inc.
Typically used for
Hosted DuckDB analytics, shared databases across a small team, cloud storage for notebook work, lightweight warehouse workloads

The honest version

MotherDuck pros and cons

No technology is all upside. This is where MotherDuck earns its place and where it tends to cause trouble.

Where MotherDuck is strong

  • Hybrid local and cloud executionA single SQL statement can read a Parquet file on your laptop and a table in the cloud, with the planner deciding where each piece runs. For notebook work this is genuinely new.
  • DuckDB SQL out of the boxAnything that runs in local DuckDB runs the same way in MotherDuck, which means existing scripts, notebooks and dbt projects move over without rewrites.
  • Priced for small teamsCompute-based pricing with a generous free tier and no per-seat charges makes it reasonable for a team of five to have a shared analytics database without a Snowflake invoice.
  • Shared databases and share linksAttaching a shared database from a notebook or sharing a read-only link with a colleague is a one-line operation, which closes a lot of distance between analyst and engineer.
  • Works from a laptop at full speedBecause the local DuckDB does real work, interactive queries on local Parquet are as fast as native DuckDB. The cloud is for sharing and durability, not for every query.
  • Good integration with modern data stackdbt, Fivetran, Airbyte, Dagster, Prefect and the usual Python notebook stack all speak MotherDuck via the DuckDB connector, so plugging it in is not a research project.

Where MotherDuck bites

  • Hybrid planner can surprise youA query that looked cheap because the data was mostly local can push a large join to the cloud and rack up compute. Reading the plan matters more here than in a pure cloud warehouse.
  • Cloud compute limits bite at scaleThe free tier and lower paid tiers have monthly compute ceilings. A heavy backfill or an accidental CROSS JOIN can burn through a month's budget in an afternoon.
  • Still a young serviceFeatures, pricing and limits have evolved quickly. A project that depends on a specific behavior should check the release notes rather than assume stability.
  • Vendor concentration is realMotherDuck is a single company running the hosted side. The engine is open, but the sharing, auth and cloud catalog are not drop-in replaceable.
  • Auth tokens are short-livedTokens expire more aggressively than some teams expect, which can surprise long-running notebooks and scheduled jobs that cached the connection on startup.
  • Not a replacement for a lakehouse at terabyte scaleMotherDuck is positioned for the small and midsize range, not for Databricks or Snowflake workloads. Pretending it is leads to bills and plans that do not match the use case.

Sound familiar?

Common MotherDuck problems we fix

The MotherDuck problems that bring people to us, and what you have once each one is fixed.

Authentication failed. Token has expired

Scheduled jobs start failing overnight

An auth token that expired during a long batch run or a notebook kernel left open across days, where the connection was cached and never refreshed. The symptom is sudden, the fix is renewal and shorter-lived connections.

Tokens refreshed on a schedule

Hybrid query: pushed to cloud (local would have been faster)

A query goes to the cloud when it should stay local

The planner decided to run work in the cloud because of a data location hint or a join that touched cloud tables, when the local path would have been faster. Reading the plan and sometimes hinting the execution brings it back.

Plans run where they belong

Query failed: too many rows read from cloud

A query blows the cloud read ceiling

A SELECT * that scanned a huge cloud table, or a join that pulled far more rows than needed. The ceiling is a guardrail; the fix is projection, predicate pushdown and better join planning.

Reads sized to the question

Local DuckDB version 1.0.0, MotherDuck expects 1.1.x

Local and hosted DuckDB versions drift

A pipeline using an older DuckDB client that no longer matches the version MotherDuck expects for extensions and features. The connection works but new features fail with cryptic errors.

Client versions pinned and current

Permission denied: user does not have access

Shared databases deny access unexpectedly

A share that was set up through the UI and never put in code, so a new user or a new team never got the grant they needed. The error does not explain which permission is missing.

Shares managed in code

Attach took 4.7s for database 'analytics'

Notebook startup is slow because attach is slow

A notebook that attaches multiple cloud databases on every kernel start, when it only needs one or two for the current analysis. The first query is slow for reasons the SQL cannot explain.

Attach narrowed to the needed databases

Catalog Error: Table 'main.events' not found

A shared view disagrees with the underlying table

A share that references a table name that moved or was renamed, so the view compiles locally but fails when the shared side tries to resolve it.

Shared views and tables aligned

Cloud compute: 1,240 seconds (limit: 10,000)

Cost headroom disappears mid-month

A new dashboard or a chatty notebook that increased cloud compute usage significantly, without the owner realizing until the warning fires near the cap.

Compute attributed and alerted on

What we do

MotherDuck help, from one slow query to a hybrid setup

Call us for one failing share link, or hand us the whole MotherDuck setup.

Rescue
Auth that keeps expiring, hybrid queries going to the wrong side, or a cloud limit about to bite. We diagnose from the plan and the usage dashboard first.
Hybrid query design
Reading EXPLAIN plans, deciding what should live locally and what should live in the cloud, and writing queries that keep the right work on the right side.
Share and permission model
Shared databases, share links, read-only grants and audit trails worked out so a growing team does not step on each other or leak data by accident.
dbt and pipeline integration
dbt-duckdb and dbt-motherduck projects set up with the right adapter, with CI runs that respect the cloud compute budget rather than burning through it.
Cost monitoring
Per-user and per-job cost attribution, budget alerts, and a review of the heaviest queries so a surprise bill does not land on a Friday afternoon.
Managed Tech
Ongoing care of the account: token rotation, version alignment, share audit, usage trends and the DuckDB upgrade path, plus a line to call when a share link goes sideways.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running MotherDuck alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

MotherDuck questions

Something else? Call+1 737 727 7327

When should we pick MotherDuck over plain DuckDB?

Plain DuckDB is right when the work lives on one machine and does not need to be shared. MotherDuck earns its keep when a small team needs a shared analytical database, cloud storage for Parquet, and the ability to query across local and hosted data in the same statement. Many teams use both, with DuckDB for development and MotherDuck for shared work.

Can MotherDuck replace Snowflake or BigQuery?

For small and midsize analytics workloads, often yes. The pricing model is friendlier at the lower end, dbt projects move over cleanly, and the hybrid model is a real advantage for notebook-heavy teams. At terabyte scale and with hundreds of users, Snowflake and BigQuery still win on operational maturity. We review the workload before recommending either direction.

How does the hybrid query planner decide where to run work?

It considers where the data lives, the size of each side, and the cost of moving rows across the boundary. Usually sensible, sometimes surprising. For queries that matter, reading EXPLAIN and occasionally hinting with pragmas keeps the planner honest.

How do we keep MotherDuck costs under control?

Attribute compute by user and job, put the heaviest queries behind reviewed changes rather than ad hoc runs, use projection and predicate pushdown aggressively, and set budget alerts before the hard ceiling. The dashboard gives you everything you need; the discipline is what teams have to add.

Can we use MotherDuck with Cloudflare R2 or S3 directly?

Yes, through the httpfs extension the same way DuckDB does locally. The usual pattern is raw data in object storage, Parquet files curated into a MotherDuck database, and queries that span both. We help with the layout decisions when a project is just starting.

How much does MotherDuck help cost?

The first review is free. Share the plan output or the error you are hitting, and we come back with a price before any work is scheduled.

Free estimate

MotherDuck giving you trouble?

Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.