01Available
Batch SQL
Run DataFusion-backed SQL and DataFrame plans over registered Arrow and Parquet-style sources.
- Finite inputs
- Arrow results
- SQL + DataFrame
Developer preview · Apache 2.0
Krishiv Engine
Compute for data that doesn’t stand still.
A Rust-native compute framework for batch SQL and stateful streaming, with experimental incremental view maintenance on one Arrow data model.
Pre-release and not for production use. Pin a commit when evaluating.
query.pyPython API
import krishiv as ks
session = ks.Session()
result = session.sql(
"SELECT 42 AS answer"
).collect()
print(result.pretty())SQLDataFusion planArrow batches
Workload model
Three shapes. One execution spine.
Shared primitives reduce the seams between finite queries, long-lived pipelines, and maintained results.
02Preview
Stateful streaming
Build event-time windows, watermarks, stateful operators, checkpoints, and continuous joins.
- Event time
- Keyed state
- Long-lived jobs
03Experimental
Incremental views
Propagate weighted inserts and retractions to maintain changing results, with visible fallback paths.
- DeltaBatch
- IncrementalFlow
- Local-first
Architecture
Move the work without changing the front door.
SQL, Rust, Python, Flight SQL, and MCP enter through public session and runtime seams. Placement remains explicit at every topology.
Read the architectureSQLRustPythonFlight SQLMCP
↓
Session + catalogDataFusion + Krishiv plans
↓
EmbeddedSingle nodeDistributed preview
↓
OperatorsShuffleStateConnectors
Built for ownership
The engine is a product on its own.
Run it inside an application, on one host, or against an explicit remote coordinator. Platform is not required and never gets a private engine path.
- Planning and execution
- Streaming operators and state
- Shuffle and checkpoint primitives
- Connector contracts and data movement
- Cross-job workflow orchestration
- Managed warehouses or billing
- Enterprise catalog administration
- Notebooks or model serving
Developer preview
Start with a verified, source-based quickstart.
No fictional package commands and no hidden production claims.