Compare · free editions only
Free databases, compared honestly.
Databases in columns, capabilities in rows. Only self-hostable free and open-source editions. No enterprise SKUs, no “contact sales”, no features that quietly require a paid plan.
How to read this table
Free means free to self-host
No license fee. You still pay for servers, backups, monitoring, and people.
Community ≠ managed cloud
MongoDB Community does not include Atlas Search or Atlas Vector Search. Those are separate paid/cloud products.
External means another component
SQLite replication, PostgreSQL vector search, and SQLite encryption depend on additional free tools or extensions.
Built for different jobs
Different databases optimize for different constraints; the table shows where each one fits.
Free-edition comparison
The table is intentionally about the free edition, not the vendor’s best paid product. A missing feature may exist in a commercial tier; it is not counted here.
| Feature | NonoDB MIT · self-hosted | MongoDB Community Server | CouchDB Apache CouchDB | SQLite Public domain | PocketBase MIT · self-hosted | PostgreSQL JSONB · PostgreSQL License |
|---|---|---|---|---|---|---|
| License | ✅ MIT | ⚠️ SSPL | ✅ Apache 2.0 | ✅ Public domain | ✅ MIT | ✅ PostgreSQL License |
| Data model | JSON document | BSON document | JSON document | Relational + JSON1 | SQLite records | Relational + JSONB |
| Deploy | Single daemon / server | Server / replica set | Single node or cluster | Embedded library | Single binary | Server / cluster |
| Query API | HTTP/JSON · Mongo wire subset · SQL | MongoDB wire | HTTP/JSON | SQL | REST + realtime | SQL |
| MongoDB wire | ✅ subset | ✅ full Community | ❌ | ❌ | ❌ | ❌ |
| SQL | ✅ | ❌ | ❌ | ✅ | ❌ | ✅ |
| Text search: engine | ✅ BM25 | ⚠️ text index | ✅ Lucene | ✅ FTS5 | ⚠️ contains/filter | ✅ tsvector/tsquery |
| Text search: ranking | ✅ BM25 score | ⚠️ textScore | ✅ Lucene score | ✅ bm25() | ❌ no relevance | ✅ ts_rank |
| Text search: fuzzy | ✅ fuzzy index | ❌ Community | ⚠️ Lucene syntax | ⚠️ trigram/extension | ❌ | ✅ pg_trgm |
| Text search: phrases | ✅ quoted phrase | ✅ quoted phrase | ✅ quoted phrase | ✅ FTS5 phrase | ⚠️ string contains | ✅ phraseto_tsquery |
| Vector search: native | ✅ native index | ❌ Community | ❌ | ⚠️ extension | ❌ | ⚠️ pgvector extension |
| Vector search: method | ✅ exact top-K | ⚠️ Atlas ANN only | ❌ | ⚠️ exact / ANN via extension | ❌ | ✅ exact / HNSW / IVFFlat |
| Vector search: metrics | ✅ cosine / dot / l2 | ⚠️ Atlas metrics | ❌ | ⚠️ extension metrics | ❌ | ✅ cosine / L2 / inner product |
| Geo indexes | ✅ GeoJSON point | ✅ 2dsphere/2d | ❌ | ⚠️ SpatiaLite extension | ❌ | ⚠️ PostGIS extension |
| Auto indexes | ✅ opt-in by query shape | ❌ | ❌ | ❌ persistent | ❌ | ❌ |
| TTL per document | ✅ _ttl field | ✅ expireAt + TTL index | ❌ | ❌ manual | ❌ | ❌ manual |
| TTL for collection | ✅ collection default | ✅ TTL index | ❌ | ❌ manual | ❌ | ⚠️ pg_cron/triggers |
| GUI | ✅ built-in | ⚠️ Compass separate | ✅ Fauxton | ⚠️ external | ✅ built-in admin | ⚠️ pgAdmin separate |
| History / audit | ✅ per-collection history | ❌ no built-in | ⚠️ document revisions | ❌ | ❌ | ⚠️ triggers/extensions |
| Replication | ✅ primary/replica/standby | ✅ replica sets | ✅ multi-master | ⚠️ external tools | ❌ no built-in | ✅ streaming/logical |
| Multi-database | ✅ one daemon | ✅ | ✅ | ⚠️ attached files | ❌ single app DB | ✅ |
| Scoped auth / tokens | ✅ principals + DB/collection scopes | ✅ roles/users | ⚠️ per-DB admins/members | ❌ no auth | ✅ auth rules | ✅ roles/RLS |
| Events / outbox | ✅ transactional outbox | ⚠️ change streams + transactions | ✅ _changes feed | ❌ | ⚠️ realtime only | ⚠️ LISTEN/NOTIFY / logical decoding |
| Streams / live updates | ✅ watch + live subscriptions | ✅ change streams (replica set) | ✅ continuous _changes | ❌ | ✅ realtime subscriptions | ⚠️ LISTEN/NOTIFY (not durable) |
| Jobs / schedulers | ✅ durable jobs/workers/schedulers | ⚠️ external queue/cron | ❌ external | ❌ external | ⚠️ app-level hooks | ⚠️ pg_cron extension |
| Encryption / tenant keys | ✅ TLS/mTLS · vault | ⚠️ at-rest in Enterprise/Atlas | ⚠️ TLS/auth · no vault | ⚠️ SQLCipher external | ⚠️ TLS proxy · no vault | ⚠️ TLS/RLS/pgcrypto · TDE external |
| Backup | ✅ native/restic · encrypted | ⚠️ mongodump/filesystem | ⚠️ replication/filesystem | ✅ file/.backup | ✅ built-in backups | ✅ pg_dump/pg_basebackup |
| Incremental / deduplicated backup | ✅ restic engine | ❌ Community | ⚠️ replication-based sync | ⚠️ Litestream external | ❌ full backups | ✅ WAL/PITR |
| Hot backup (online) | ✅ no daemon stop | ✅ mongodump/snapshot | ✅ replication/snapshot | ✅ backup API | ✅ built-in backups | ✅ pg_dump/pg_basebackup |
| Backup scheduler | ✅ built-in scheduler | ❌ external cron | ❌ external | ❌ external | ✅ built-in backups | ❌ external |
| Encrypted backups | ✅ gzip/age · native/restic | ❌ external | ❌ external | ❌ external | ❌ unencrypted | ⚠️ external gpg |
| Best fit | Document apps consolidating storage, search, SQL, and automation | Existing MongoDB ecosystem | Offline-first sync | Embedded/local-first | Small apps/admin panels | General-purpose SQL |
Text search is not one feature
A lexical engine, relevance ranking, fuzzy matching, phrase support, and filtering are separate capabilities. MongoDB Community has text indexes, but Atlas Search is a different paid/cloud product. CouchDB uses Lucene; SQLite uses FTS5; PostgreSQL uses tsvector/tsquery.
Vector search is not one feature
Check whether vectors are native or an extension, whether the search is exact or ANN, and which distance metrics are supported. NonoDB is native and exact; PostgreSQL needs pgvector; MongoDB Community has no vector search outside Atlas.
This is a product-level comparison, not a benchmark. Free tiers and editions change; verify licensing and feature gates before committing. “No built-in” does not always mean impossible — it may require an external free tool.
Footprint for 1M documents
Baseline: roughly 1 KB per document, three simple secondary indexes, and no text or vector indexes. Only NonoDB numbers are measured from the published benchmark. The other rows are planning ranges, not apples-to-apples measurements.
| Database | Steady RAM | Disk for 1M docs | Notes |
|---|---|---|---|
| NonoDB | ~5–10 MB idle / after reads | ~206 MB base · ~445–465 MB with 3 indexes | Measured on ext4. Full scan + sort: ~278 MB RSS; peak parallel backfill: ~552 MB. 100k text docs + search indexes: ~627 MB. |
| MongoDB Community | 256–512 MB+ | 0.5–1.5 GB | WiredTiger; replica sets and Atlas add overhead. Community has no Atlas Search/Vector. |
| CouchDB | 100–300 MB | 0.5–2 GB | Erlang VM; cluster and Lucene indexes add memory and disk. |
| SQLite | 5–30 MB | 0.1–0.5 GB | In-process; page cache is usually counted in the application, not the database. |
| PocketBase | 20–80 MB | 0.1–0.5 GB | SQLite plus Go runtime; single node, no cluster overhead. |
| PostgreSQL + JSONB | 50–200 MB | 0.3–1.5 GB | Depends on shared_buffers, WAL, autovacuum, and extensions such as pgvector. |
NonoDB measured
Idle after restart with 1M documents and three indexes: ~5.1 MB RSS. After a few reads: ~6.4 MB. Disk: ~206 MB before secondary indexes, ~445–465 MB after three indexes.
What changes the numbers
Document size, index count, text/fuzzy/vector indexes, history retention, page cache, replication, and backup retention can move disk and RAM by multiples. Benchmark with your own data.
NonoDB figures come from the published 1M-document benchmark on ext4. Other ranges are order-of-magnitude planning estimates for small operational documents; they are not direct measurements on the same dataset. See the NonoDB benchmark methodology.
Which free database should you pick?
| Choose | When it is the better fit |
|---|---|
| NonoDB | You want one binary with documents, search, SQL, history, TTL, auto indexes, backups, replication, auth, and a GUI. |
| MongoDB Community | Your team already knows MongoDB and needs the broadest driver/tool ecosystem. |
| CouchDB | Offline-first sync and multi-master replication matter more than SQL or Mongo compatibility. |
| SQLite | The database lives inside one process and does not need a network server. |
| PocketBase | You want a small backend with auth, realtime, backups, and an admin UI for a prototype or small app. |
| PostgreSQL + JSONB | You need a proven general-purpose SQL database and JSON documents are only one part of the product. |
Where NonoDB deliberately differs
- One binary, not a stack: documents, search, SQL, history, TTL, events, backups, replication, auth, GUI, and auto indexes ship together.
- Auto indexes are real: opt-in creation for frequent equality/range query shapes, with unused automatic indexes retired. Manual indexes are never dropped.
- TTL at both levels: a collection default and a per-document
_ttloverride. - MongoDB-compatible subset: official drivers work inside the documented subset; unsupported operations fail loudly instead of pretending.
- No paid feature gates: the MIT binary does not hide search, backups, replication, or the GUI behind an enterprise tier.
Compare capabilities, not marketing.