Serverless FerretDB

FerretDB is an Apache 2.0 licensed database that implements the MongoDB wire protocol on top of PostgreSQL, so MongoDB drivers and mongosh connect to it without code changes. It exists so document workloads can have Mongo ergonomics without the SSPL that covers MongoDB itself. Layerbase runs it as a managed endpoint with TLS, automated backups, a document console, and copy-on-write branching, which comes free because the documents live in Postgres underneath.

A `mongodb://` endpoint with TLS, automated backups, IP allowlisting, and a document console in the browser, plus hibernation that scales the database to zero when it is idle and wakes it on the next connection. There is no cluster to size and no cloud provider account to link: sign up, create the database, copy the connection string.

FerretDB implements the MongoDB wire protocol on top of PostgreSQL. From your driver, it looks like Mongo. Under the hood, your documents are stored in PostgreSQL through the DocumentDB extension (a binary BSON type), not plain tables you query directly. Licensing: FerretDB is Apache 2.0 and the DocumentDB extension is MIT, instead of SSPL.

Yes. Any driver that speaks the MongoDB wire protocol (pymongo, mongoose, mongo-go-driver, Java, etc.) connects to FerretDB without code changes. mongosh works against the same endpoint.

The common 90% is covered: CRUD, indexes, aggregations, and transactions. Mongo-only features like change streams and Atlas Search are not supported. The FerretDB compatibility page tracks specifics.

FerretDB is not on the Free plan. It starts on Solo at $5/mo, with Pro at $15/mo above it; both are shared plans that hibernate to zero when nothing is connected, so idle time costs nothing. Above those sit flat-rate dedicated presets at $35/mo (2 vCPU / 4 GB), $65/mo (4 vCPU / 8 GB), and $120/mo (2 vCPU / 16 GB) with storage included. Compared on what you pay for the hardware, Atlas compute alone lists higher: verified 2026-08-19 from mongodb.com/pricing and converted at 730 hours per month, M10 (2 vCPU / 2 GB) is about $58, M20 (2 vCPU / 4 GB) about $146, and M30 (2 vCPU / 8 GB) about $394, with storage, backup, and egress billed on top. That is a price-for-hardware comparison and not a performance one: we have never run a benchmark against Atlas. Worth knowing before you switch: an Atlas M-tier is a three-node replica set, where a Layerbase dedicated preset is a single node with backups and branching.

We benchmarked FerretDB against self-hosted MongoDB 8.0 with YCSB, both on the same hardware profile: the box behind our $65/mo dedicated preset, on a working set larger than RAM. FerretDB on our tuned default did 3.3-3.8x MongoDB on a 50/50 read/write workload and about 5x on reads. The tail latency is the stronger result: on those same runs, tuned FerretDB held update p99 between 0.1 and 0.85 seconds across every thread count we tested, where MongoDB ran 0.16 to 1.7 seconds, so roughly half the tail at each level of concurrency. Those numbers come with real caveats (one hardware profile, MongoDB self-hosted on that same hardware rather than Atlas, so no Atlas run stands behind any of it, and MongoDB wins at much larger scales), so read the full writeup rather than taking the multiples at face value.

Three questions decide it. First, working set versus RAM, which dominates everything else: if your working set fits in RAM, expect parity with MongoDB rather than multiples, and the numbers above were measured with a working set larger than RAM on one reasonable box. Second, absolute scale: at 100 million to a billion or more documents on a write-heavy workload, MongoDB genuinely wins, and we would rather say so than sell you the wrong thing. Third, your read/write mix: reads favored FerretDB in every configuration we ran. One honest note about the ladder: the $35/mo preset is sized for a working set that fits in 4 GB, so hand it a 10 GB working set and it is the slowest thing we sell. If your workload shape is not represented in the writeup, tell us what it looks like and we will run it on our hardware at our cost and send you the numbers, including the ones that do not flatter us.

Yes, and this is the capability Atlas does not offer at any tier. Because a FerretDB database on Layerbase is a Postgres database wearing a Mongo interface, we branch it copy-on-write with the same filesystem mechanics we use for Postgres: hit Branch, get an isolated copy of the data in seconds, point a preview deploy or a CI job or an agent at it, and delete it when the pull request merges. No export, no restore, no snapshot to schedule, no Mongo-specific machinery. The closest thing Atlas has is a snapshot restore, into the same cluster or a different one, and a restore is a slow full copy of the data rather than a branch you take per pull request.

Yes. mongodump and mongorestore work against FerretDB. Most apps move with zero driver-side changes; you typically only re-verify queries that exercise less-common operators.

Yes. Paste your mongodb+srv:// string and Layerbase copies the collections, documents, and indexes into a new FerretDB database for you, reading Atlas once and never writing to it. A MongoDB you host yourself works the same way, as long as the server is reachable from the public internet.

/db/mongodb explains the Mongo story on Layerbase Cloud, which is implemented by FerretDB. /db/ferretdb is the same engine viewed through its own brand and license story; pick whichever framing fits your audience.