Notes

Compare self-hosted Instant with Datahike's thin, replicated and embedded clients, including permissions, optimistic writes, history and operating costs.

Why Datahike's serializer is plain CBOR: a standard format at the speed of the fastest private ones, readers in other languages, and reads into stored values without decoding them.

How purge, garbage collection and access control work across Datahike, including the storage layers they do not cover.

How Datahike creates branches with a small set of konserve writes and uses multi-source Datalog to compare and merge them.

How the pg-datahike beta exposes Datahike through the PostgreSQL wire protocol for psql, migrations and supported ORM workflows.

Why Stratum runs isolation forests inside its SQL engine, what that removes from a pipeline, and where the current implementation fits.

How immutable snapshots, copy-on-write branches and coordinated versions can support reproducibility, scenario analysis and regulated-data workflows.

How independent Datahike processes share immutable indexes through storage and join database values without a coordinating server.

How Stratum combines a JVM columnar SQL engine, SIMD execution and copy-on-write branches, with benchmarks and implementation trade-offs.

A personal account of how functional values, persistent data structures and long-lived systems led to Datahike.

How Yggdrasil defines shared snapshot, branch, merge and history operations across different storage systems.

How copy-on-write storage, structural sharing and named branches apply version-control ideas to database values.

How immutable vector and full-text indexes make retrieval reproducible across historical and branched states.