Kaveon Engine

Why it exists

Most analytics products wrap an engine someone else wrote — DuckDB embedded, Trino deployed alongside, or SQL pushed down to whatever the customer already runs. Kaveon Engine is first-party so that execution, storage access, and the semantic layer can be designed against each other rather than negotiated across a boundary. It embeds no other engine.

Install

# Linux x64 / Apple Silicon macOS — engine-dev preview, installs to ~/.local/bin
curl -fsSL https://raw.githubusercontent.com/PruthviProdduturi/Kaveon/dev/scripts/install.sh | bash

# A tagged release (adds Intel macOS), verified against its SHA256SUMS
curl -fsSL https://raw.githubusercontent.com/PruthviProdduturi/Kaveon/dev/scripts/install.sh | KAVEON_VERSION=0.3.0 bash

# Windows x64 (PowerShell; no source clone required); set $env:KAVEON_VERSION for a release
irm https://raw.githubusercontent.com/PruthviProdduturi/Kaveon/dev/scripts/install.ps1 | iex

# From source
cd engine && cargo install --path crates/cli

Tagged releases are listed at github.com/PruthviProdduturi/Kaveon/releases with one archive per platform and a SHA256SUMS file the install scripts verify against. Each release also attaches a winget manifest and a Homebrew formula for whenever a package-manager listing is wanted; the scripts and the archives are the supported installs.

Define a catalog

A catalog points the Engine at storage. Drop one TOML file per catalog into ~/.kaveon/catalogs/ — the filename becomes the catalog name — or declare them inline in ~/.kaveon/config.toml.

# ~/.kaveon/catalogs/warehouse.toml
type      = "local"
base_path = "/data/warehouse"

Tables are auto-discovered from the .parquet files in base_path, and Delta tables from directories carrying a complete JSON commit log. Register tables explicitly when you want to control the name, schema, or format:

[[table]]
name     = "orders"
schema   = "default"
location = "orders/"
format   = "delta"       # parquet | delta | iceberg
access   = "shortcut"    # read in place, no rewrite

Run a query

The installed client connects to a coordinator by default. With Microsoft authentication enabled,--auth auto reuses Azure CLI login when possible, then falls back to Microsoft device sign-in. Use --auth azure-cli to require Azure CLI or --auth microsoft to choose device sign-in. The UI identifies the client as Kaveon CLI and shows the signed Entra username; immutable Entra object identity remains the ownership key.

kaveon --server https://engine.example.com --catalog medallion --schema test

# AKS port-forward with the deployment CA
kubectl -n kaveon port-forward service/kaveon 18443:8080 --address 127.0.0.1
kaveon --server https://localhost:18443 --ca-cert ./kaveon-ca.crt --catalog medallion --schema test

See the Azure deployment guide for certificate handling and the full connection procedure.

The CLI runs embedded with --local, which needs no server:

kaveon --local --data-dir /data/warehouse

SHOW CATALOGS;
SHOW TABLES;
SELECT region, count(*) AS n
FROM   orders
GROUP  BY region
ORDER  BY n DESC
LIMIT  5;

Or execute one statement and exit — useful in scripts:

kaveon --local --data-dir /data/warehouse -e "SELECT count(*) FROM orders"

Interactive shell and scripts

CLI 0.2.0 adds persistent history, reverse history search, tab completion, and EMACS or VI editing. The prompt shows your active schema in white with a muted gray prefix; NO_COLOR disables terminal colors. Use exit or quit to leave the shell.

# Execute a script against the connected Engine
kaveon https://localhost:18443/medallion/test --ca-cert ./kaveon-ca.crt --file queries.sql

# Export one JSON object per row
kaveon https://localhost:18443/medallion/test --ca-cert ./kaveon-ca.crt --execute "SELECT * FROM orders LIMIT 10" --output-format JSON

Remote scripts and redirected standard input support multiple statements. A failed statement stops the batch; --ignore-errors continues while retaining a nonzero exit status. Uppercase output formats include ALIGNED, AUTO, VERTICAL, MARKDOWN, CSV_HEADER, TSV_HEADER, and JSON. Existing lowercase json retains its array-of-rows format.

Save connection defaults as key=value lines in ~/.kaveon_config, or select a file with KAVEON_CONFIG. See the CLI guide for all formats and examples, and the compatibility checkpoint for the remaining differences from Trino.

Run the cluster

For distributed execution, start a coordinator and one or more workers, then point the CLI at the coordinator. Without --local the CLI is a thin remote client and defaults to http://localhost:8080.

cargo run -p kaveon-server -- /etc/kaveon/coordinator.toml
kaveon --server http://localhost:8080

The Compose stack in the Quickstart does this for you — a coordinator plus two workers, with the coordinator published on port 8081.

Environment variables override TOML. Operational endpoints: /ui for the console, /health and /ready for probes.

What SQL runs today

HTTP surface

curl -s localhost:8080/v1/statement \
  -H 'content-type: application/json' \
  -d '{"query":"SELECT count(*) FROM warehouse.default.orders"}'

Read next

• Architecture — processes, crates, and startup.
• Engine SQL — semantics and integration gates.
• Distributed Runtime — stages, fragments, exchange, retry.
• Storage & Catalogs — reads and native metadata.
• Engine Memory — reservations, spill, and safety boundaries.