At a glance
| Supported databases | PostgreSQL and Amazon RDS (agent README); other engines not confirmed on surviving pages |
| Query analysis | Agent collected query and long-running query data; service features no longer documented |
| Execution plans | Not documented |
| Wait statistics | Not documented |
| Index recommendations | Not documented on surviving official pages |
| Slow-query detection | Long-running query monitoring flag in the agent |
| Tuning automation | Machine-learning knob tuning, the core idea |
| Deployment | Discontinued SaaS; archived research code |
| Licensing | Research code BSL 1.1 to Apache 2.0 (2024); agent Apache-2.0 |
OtterTune is no longer an option to buy. It is still worth knowing as the best-known example of automatic knob tuning with machine learning, and its research code can be studied, but for a live system pick a maintained monitoring or tuning tool for your engine.
Overview
OtterTune started as a Carnegie Mellon Database Group research project (the project page lists 2013 to 2021) by Dana Van Aken, Andy Pavlo and Geoff Gordon. It used machine learning to choose values for the hundreds of configuration knobs a database exposes, such as memory sizes and write frequency, by mapping a workload to previously seen ones and reusing what earlier tuning sessions had learned. The work was published at SIGMOD 2017 and VLDB 2021.
Pavlo co-founded OtterTune as a company to sell the idea as a cloud service. Its public GitHub repositories show an agent that collected metrics, index statistics, query and schema information from databases, with AWS IAM authentication for Amazon RDS, plus Terraform, CloudFormation and Helm templates for customers. The agent was archived on 14 June 2024 and the company closed that year; Pavlo's own page now says the start-up "died".
Key features
- Machine-learning recommendations for database configuration knobs (the research tool and later service)
- Reuse of knowledge from earlier tuning sessions to cut training time for a new workload (CMU project description)
- Agent that collected database metrics, index and table statistics, query and schema information (archived ot-agent)
- AWS IAM authentication for Amazon RDS in the agent
- Public API used to apply knob changes through Terraform and GitHub Actions (archived example repository)
Integrations and ecosystem
Pros and cons
Pros
- Well-documented research on machine-learning configuration tuning, with peer-reviewed papers
- Source code remains public for study under Apache 2.0 terms
- Useful reference when evaluating other automatic tuning products
Cons
- The company and its service shut down in 2024; there is no support or service to buy
- Repositories are archived and receive no fixes
- Former product, documentation and pricing pages are offline, so many details cannot be confirmed
Pricing
There is nothing to buy: the OtterTune company shut down in 2024 and its commercial tuning service is gone. What remains is archived source code. The original CMU research tool (cmu-db/ottertune) carries a Business Source License 1.1 whose change date of 18 January 2024 converts it to the Apache License 2.0, and the company's metric-collection agent (ottertune/ot-agent) is Apache-2.0. Both repositories are read-only and unmaintained. The commercial service's former prices are not available from any surviving official page and are not given here.
Pricing checked on the vendor's official page on 7 November 2026. Prices change; confirm before buying.
Who it is best for
Researchers and engineers studying automatic database configuration tuning, or anyone reading older material that recommends OtterTune.
Typical use cases
- Studying how machine-learning knob tuning works, from the CMU papers and code
- Understanding what a former OtterTune deployment collected before replacing it
- Benchmarking ideas for configuration tuning in a research setting
Who should look elsewhere
- Anyone who needs a supported tuning or monitoring service today: see PostgreSQL monitoring tools or MySQL monitoring tools.
Verdict
OtterTune helped popularise machine-learning database tuning, but the company shut down in 2024 and nothing remains to buy or run in production. Read its research if the topic interests you, and use a maintained tool for real systems.
Frequently asked questions
Is OtterTune still available?
No. The company shut down in 2024. Co-founder Andy Pavlo describes the start-up as dead on his Carnegie Mellon page and in his retrospective dated 1 January 2025, the agent repository was archived on 14 June 2024, and ottertune.com no longer resolves.
Can I still run OtterTune myself?
The original CMU research code is public but archived since March 2021. Its Business Source License 1.1 converted to the Apache License 2.0 on 18 January 2024. It is a research prototype with no maintenance, so it is not a practical production tool.
What did OtterTune tune?
Database configuration knobs, such as memory and write settings, chosen by machine-learning models trained on earlier tuning sessions. The CMU project focused on knob configuration rather than rewriting individual queries.
What should former OtterTune users use instead?
For PostgreSQL, pganalyze or Percona Monitoring and Management give query and configuration insight; Datadog Database Monitoring covers several engines. None is a drop-in copy of OtterTune's machine-learning knob tuning.
Sources
- CMU Database Group: OtterTune project
- Andy Pavlo, Carnegie Mellon home page
- Andy Pavlo: Databases in 2024, a year in review (1 January 2025)
- cmu-db/ottertune repository (archived)
- cmu-db/ottertune licence (BSL 1.1)
- ottertune/ot-agent repository (archived 14 June 2024)
Checked November 2026.
Listed in: Performance & Query Optimization.
How we research reviews and what star ratings mean: our editorial method.