Kaveon vs HTAP platforms
How Kaveon’s unified transactional and analytical direction compares with Snowflake Unistore, Databricks LTAP, TiDB X, and Microsoft Fabric.
Market map
| Platform | Engines | Durable authority | Open source | Status |
|---|---|---|---|---|
| Snowflake Unistore | One platform, Hybrid Tables | Proprietary format on object storage | No | GA November 2024 on AWS |
| Databricks LTAP | Two: Postgres (Lakebase) and Spark/Photon | Object storage, Delta and Iceberg | Partial | Announced June 2026, “coming soon” |
| TiDB X | One | Object storage as single source of truth | Apache 2.0 | Announced October 2025; GA unconfirmed |
| Fabric SQL database | SQL plus Spark | OneLake, read-only Delta mirror for analytics | No | GA |
| Kaveon | One (target) | ADLS Gen2 (target) | MIT | Analytics alpha; transactional layer not implemented |
Where Kaveon differs
Data is queried in place. Kaveon reads customer Parquet and Delta files where they already reside and registers them through the catalog. Snowflake and TiDB require data to be loaded into engine-owned storage; Fabric mirrors into OneLake. For data already in ADLS Gen2, this is the difference between a registration and a migration.
Natural language is deterministic. The DLM resolves questions to SQL and precomputed context with no hosted model, so the same question yields the same answer and the derivation is auditable. No platform surveyed provides a deterministic natural-language interface, and TiDB provides no semantic layer at all.
Maturity, stated honestly
Kaveon is behind every platform in this comparison on transactional maturity. TiDB has approximately a decade of production operation and a CNCF Graduated storage engine; Snowflake Unistore has been generally available for close to two years. Kaveon has a foundation: typed product records with revision-checked writes, uniqueness and reference validation, digest-verified immutable documents, and a conditional head update. Row DML, constraints, isolation, index shard splitting, verified head recovery, and the migration path remain open. PostgreSQL remains the authoritative product store.
Canonical paper
The full market map, per-platform notes, defensible claim, and primary references live in Kaveon and unified transactional/analytical platforms.