methodology
How WhichDB works
WhichDB has two jobs: measure how widely each database is used, and help you pick one by fit. The two are kept separate, and both are explained here.
The WhichDB Score (popularity)
The score is a 0–100 index built from 8 public signals. Each signal is log-scaled against a reference value near the top of the field, because adoption follows a power law: ten times the downloads means more popular, not ten times better known.
| signal | weight | what it captures |
|---|---|---|
| Client downloads | 22% | Monthly downloads of the most-used official client on npm, PyPI and crates.io: applications using it today. |
| Hacker News | 15% | Stories and comments mentioning the database in the last 12 months: current developer mindshare. |
| Wikipedia | 15% | Average monthly English Wikipedia pageviews over the last 12 months: general awareness. |
| Stack Overflow | 15% | All-time questions under the primary tag: installed base. Only counted for databases released by 2021, since the site's question volume collapsed after 2022. |
| Job posts | 10% | Posts in the last 12 Hacker News “Who is hiring?” threads that mention the database: hiring demand. |
| GitHub stars | 8% | All-time stars on the main repository: cumulative open-source interest. |
| Docker Hub | 8% | All-time pulls of the official image: real-world deployments. |
| Stars gained (12 mo) | 7% | GitHub stars gained over the last 52 weeks: open-source momentum. |
Signals are combined as a weighted mean over the signals that apply to a database. A closed-source product isn't penalised for having no GitHub repository, a managed service isn't penalised for having no Docker image, and a database without its own client library (one that speaks the PostgreSQL or MySQL protocol, say) isn't penalised on downloads. A signal that applies but is missing counts as zero. A gentle curve spreads out the top of the table. Deprecated systems are discounted by 20% and discontinued ones by 40%.
The score measures adoption and mindshare. It does not measure quality, performance or fit. A young database can be the right choice for you long before it climbs the table.
Why Stack Overflow is no longer the main signal
Until 2022, Stack Overflow questions were the best public proxy for developer usage, and most rankings still lean on them. Since AI coding assistants arrived, the site's question volume has fallen about 99% from its 2014 peak. In the twelve months to September 2026, the postgresql tag gained a few hundred questions, and many newer databases gained none.
The all-time question count still tracks installed base well, so WhichDB keeps it at a modest weight, but only for databases released by 2021. Anything newer never had a fair chance to build up questions. Most of the weight now sits on signals from the last twelve months: client downloads, Hacker News discussion, job posts and GitHub star growth. The full analysis, with measurements and the alternatives that were rejected (Reddit, Google Trends, Bluesky, GH Archive), is in research/signals-2026.md.
Smart match (fit)
When you describe a project, a deterministic parser maps phrases to explicit requirements, for example “payments” → ACID transactions as a must-have. It runs entirely in your browser. Nothing you type leaves the page, and the parser shows exactly which phrases it picked up.
Each database is then scored on each requirement with partial credit. A related workload earns about half, and a scale tier one step short earns about 45%. Must-haves weigh 1.6× and cap the result below 50% when unmet. The final fit is 88% requirement coverage and 12% popularity, a tie-breaker that stands in for ecosystem, documentation and hiring pool. Research prototypes and discontinued systems are left out unless you ask for them.
Data & freshness
Every profile is a JSON file in the public repository. Profiles are researched from official documentation, release notes, repositories, papers and vendor announcements, and each lists its sources and the date it was last verified. The 844 profiles were verified between Oct 27, 2026 and Oct 27, 2026. The catalogue covers active databases, managed services, extensions and embedded libraries, plus the discontinued products and research systems people still search for, each clearly marked with its successors.
Popularity signals are collected by a script from public APIs (GitHub, npm, PyPI (via ClickHouse's public PyPI dataset and pypistats), crates.io, Hacker News via Algolia, Wikimedia, Docker Hub and Stack Exchange). They are refreshed monthly, and monthly snapshots power the trend column in the rankings.
The catalogue is also machine-readable: /data/index.json (every database), /data/db/<slug>.json (one full profile) and /llms.txt (a summary for AI assistants).
Contribute
Spotted something outdated? Every database page has a suggest an edit link that opens its JSON file on GitHub. The schema is strict, so the build checks every change. Missing databases are welcome too.