A database built for your workload, that keeps every version.
MaDB is a database for business data that keeps every version, lives on object storage and is built for each customer's workload, and is designed to keep tuning itself.
The problem
Software companies keep their customers’ business data, such as orders, invoices, accounts and documents, in a general-purpose database. As they grow, it stops being enough. Search and filtering move to a second system, analytics to a third, and an audit trail is bolted on by hand. Each copy has to be kept in sync, each one costs money even for customers who are idle, and none of them can answer a simple question: what did this data look like last Tuesday?
AI agents make this worse. An agent that changes customer data needs a record of everything it did, and a way to see and reverse its mistakes. Most databases forget the past the moment it is overwritten.
What MaDB is
Each change is a new version, and by default every version stays readable: one record, a filtered list or a total, now or at any point in the past. A shorter history is one setting away.
Data files live in Amazon S3 today, with Google Cloud Storage and Azure Blob Storage on the way. The small catalog that names each version moves to durable storage next. From then on servers hold nothing that cannot be rebuilt, so an idle customer costs only the bytes stored.
The database is generated from the customer's data model, with exactly the indexes and layouts their queries need, and nothing interpreted on hot paths.
Cheap when idle
Most business software is multi-tenant. A few customers are busy at any moment, and most are not. Databases that run on dedicated servers and disks charge for every customer all the time. MaDB stores everything in object storage, where data at rest is billed only for its size, and runs servers only for the customers in use.
A database that tunes itself
Most databases meet most of a customer’s needs and leave a few out, and the customer works around the gaps with more systems or more engineering time. MaDB is built from the customer’s own data model, and is designed to keep adapting to their workload: adding the indexes and layouts their queries need, and dropping the ones nobody uses. The goal is a promise customers can check every week: costs go down and performance goes up, without their engineers.
Why now
Since 2024, cloud object storage offers single-digit millisecond reads, conditional writes and appends: enough for online workloads, not only archives.
Rust and modern compression and bitmap research make it possible to build a fast engine with a very small team.
AI agents acting on business data need audit, safe experiments and undo, which only a versioned database gives natively.
Where it stands
The core engine is written and tested: versioned commits on object storage, reads at any version, background compaction, compression, and point reads in about 7 µs once the data is cached. It is checked by deterministic simulation against a reference model, with injected storage failures and writer restarts, on every change.
Next come a published benchmark at 100 million entities on object storage, queries across many fields, and distribution across servers.
Team
Founder name, photo and short bio.
Contact
Contact address.