--- title: "Hypertenancy: Multi-Tenant Compute Isolation | MotherDuck" description: "Hypertenancy gives each user, customer, or AI agent isolated per-tenant compute — high-concurrency, multi-tenant analytics with no shared cluster or noisy neighbors." canonical: "https://motherduck.com/product/hypertenancy/" --- HYPERTENANCY # Dedicated compute Full isolation. Give each user, customer, or AI agent a dedicated instance — fully isolated, independently sized, and scaled to zero when idle. [GET STARTED FREE](https://auth.motherduck.com/authorize?app_source=web&response_type=code&client_id=bza3KWQpxRAFlTlRFXUo29AOg9xD7zcp&redirect_uri=https%3A%2F%2Fapp.motherduck.com%2F&state=STATE&auth_flow=signup&screen_hint=signup&ext-ph_distinct_id=2a1309a0-cb8e-4e1f-850f-0da3fd1eb2ff) [VIEW DOCUMENTATION](https://motherduck.com/docs/concepts/hypertenancy/) Diagram showing an AI agent and human users routing through MotherDuck to individually sized Duckling instances (Mega, Standard, Pulse) that connect to internal and end-user databases ## Scale Vertically and Horizontally Manage DuckDB compute instances by user, then add read scaling replicas for read-heavy workloads. UI Curl Python Scaling UI > We were able to identify specific organizations, and even better, specific users, who need more compute. We can scale individually for each one, so we get a much faster UX without paying to scale an entire cluster. Josh Nakka Co-Founder @ Pricemedic Josh Nakka [Read the Case Study](/case-studies/pricemedic-hypertenancy-healthcare-transparency/) ## A More Flexible Compute Model How It Works ### One Duckling per user, scaled independently Hypertenancy allows you to manage compute instances independently for sub-second query performance with dramatically lower costs. Provision Pulse instances for your agents, speedy Standards for your most important app users, or a Mega for that backfill — you'll only pay for what you use. For Agents ### Sandboxed for AI Agents Agents can generate SQL without running amok. Hypertenancy sandboxes each agent's compute, so a runaway query can't blow up your bill or take down your warehouse. Read Scaling ### Scale out for read-heavy workloads Attach read-scaling replicas to individual compute instances to handle read-heavy workloads from agentic applications and business intelligence tools. Scale instances independently with as many replicas as you need. [READ SCALING DOCS](https://motherduck.com/docs/key-tasks/authenticating-and-connecting-to-motherduck/read-scaling/) Cost Efficiency ### Stop paying for idle resources With other warehouses, you provision for peak and pay for that capacity always. Using hypertenancy, you get a serverless architecture built for real-world workloads. [LEARN MORE ABOUT COSTS](https://motherduck.com/blog/faster-ducks/) Admin API ### Manage programmatically Create service accounts and provision compute programmatically via our REST API. Each user-internal or end-user-gets a dedicated duckling that spins up on first query and scales to zero when idle. No cluster scaling or tuning required. [ADMIN API DOCS](https://motherduck.com/docs/sql-reference/rest-api/motherduck-rest-api/) ## FAQS ### What is Hypertenancy? Hypertenancy provisions a dedicated DuckDB instance (Duckling) per user or customer, providing full compute isolation. Unlike traditional multi-tenant warehouses that share a single cluster, each tenant's queries run on isolated compute — heavy workloads from one tenant cannot impact others. Read more in the customer-facing analytics documentation. ### What sizes of Compute Instances are available? MotherDuck compute instances are called Ducklings. Each Duckling operates as a single-node instance of DuckDB, offering sub-second query performance without the network latency of distributed query engines. Ducklings are independently sized: Pulse, Standard, Jumbo, Mega, Giga. [Read more about instance types](/docs/about-motherduck/billing/duckling-sizes/). ### How do I manage Compute Instances in my application? MotherDuck provisions ducklings automatically per service account. Create a service account for each customer, generate read/write tokens, and connect using `\`md:\`` connection strings. Your app uses tokens to route queries; MotherDuck handles instance lifecycle, scaling, and spin-down. ### What is Read Scaling? Read scaling allows you to provision read replicas for use in read-heavy workloads, like concurrent use of business intelligence tools. This feature provides high availability and performance, while maintaining your ability to separately provision read/write instances. [Learn more about read scaling](/docs/key-tasks/authenticating-and-connecting-to-motherduck/read-scaling/). ### Does MotherDuck handle high-concurrency, multi-tenant queries? Yes. Hypertenancy gives every user, customer, or AI agent its own Duckling — dedicated, isolated compute — so concurrent tenants never queue behind each other or hit "noisy neighbor" slowdowns. For read-heavy tenants, Read Scaling adds replicas, so you scale concurrency without provisioning or protecting a shared cluster. ### Do I still need a caching layer for customer-facing analytics? Usually not. Because each tenant already runs on isolated per-tenant compute, there's no shared cluster to shield with a cache tier — so there's no pre-aggregation to build, refresh, or invalidate. A cache still helps for extremely high-QPS identical queries; for per-tenant distinct queries, isolated compute is the simpler design. [Learn why embedded analytics can skip the cache](/learn/skip-the-cache-embedded-analytics/). ## Bring your architecture to the Duck side Deploy sandboxed agents, built responsive applications, or simply power up your data stack with sub-second queries for every user. [Try MotherDuck Free](https://auth.motherduck.com/authorize?app_source=web&response_type=code&client_id=bza3KWQpxRAFlTlRFXUo29AOg9xD7zcp&redirect_uri=https%3A%2F%2Fapp.motherduck.com%2F&state=STATE&auth_flow=signup&screen_hint=signup&ext-ph_distinct_id=2a1309a0-cb8e-4e1f-850f-0da3fd1eb2ff) [READ THE CFA GUIDE](https://motherduck.com/docs/getting-started/customer-facing-analytics/)