.mvs dataset: a documented tree of objects in an object-storage bucket. There is no proprietary storage service in the path. The same layout runs on GCS, S3, MinIO, or your own hardware.
A .mvs dataset is a prefix, not a single file. Ranged reads, parallel restore, and lazy loading all depend on that shape.
The format has two versions. Mixpeek is moving namespaces from v1 to v2 in November 2026. Namespaces with multi-vector fields stay on v1 until v2 stores multi-vector indexes.
v2 layout
In v2, each namespace is one key prefix with no shard level. Every object is written once and never changed in place.chain/, then reads the wal/ commits written after it. It answers from both, so a committed write is visible before compaction folds it into a generation.
Vectors are stored as RaBitQ codes for the first pass, with float16 copies for reranking.
v1 layout
meta.json carries a format identifier (lire_rs_v1). Readers refuse identifiers they do not recognize, so a future format change cannot silently misread your data.
What the format guarantees
In v1:- Pairing. Every indexed point has both an id and a vector entry. A point cannot be countable but unsearchable.
- Parts sum to whole. Partition row counts add up to the manifest’s total, and the manifest’s total reconciles with the payload store.
- Manifest honesty. Every partition the manifest declares exists as an object, by name. Vanished blobs surface as findings, not as silently missing results.
- Durability frontier. The snapshot’s replay floor never exceeds the shipped write-ahead log. Recovery can always replay forward from a snapshot to the last shipped write.
- Checksums. The snapshot image carries an inline SHA-256 trailer. Restore refuses a mismatch.
- Digests on every object. Each generation lists every object it uses with its size and an xxh3-64 digest. Readers check the digest before they use the bytes.
- Create-only commits. Commits, chain entries and generation descriptors are written with a create-only check, so two writers cannot overwrite each other.
- Reuse without rewrite. A new generation lists the objects it reuses from earlier ones, so compaction does not rewrite data that did not change.
- Manifest honesty. Every object a generation lists exists, with the listed size.
Why this matters to you
No lock-in. Your data sits in your bucket in a documented layout. A tool that implements this page can read it without Mixpeek in the path. Auditability. Regulated teams can verify what is stored, where, and whether it is internally consistent, from the objects alone. Portability. Copying a namespace to another bucket is an object copy: one prefix in v2, three trees in v1. The format validates identically at the source and the destination, on any object store. Tiering without translation. Warm and cold data use the same layout at different object-storage classes. Nothing is rewritten when data moves between tiers.Status
The v1 specification is complete. The v2 specification is a draft that describes what Mixpeek writes today, and it becomes v2.0 when its open items close. Enterprise customers can request either byte-level specification.The write path stays exclusive to Mixpeek: all ingestion flows through the API and processing pipeline. The format documents what we store so you can always read, verify, and move your own data.

