CLI Commands

The CLI binary is kino. The gRPC server binary is kino-serve.

kino <COMMAND>

kino create-test

Generate synthetic .kdb files for smoke tests and benchmarks.

kino create-test [PATH] \
  -n <NUM_EPISODES> \
  --frames <FRAMES> \
  --images \
  --compress <QUALITY>

Option	Default	Description
PATH	test.kdb	Output path
-n, --num-episodes	10	Number of episodes
--frames	50	Frames per episode
--images	false	Add fake 64x64 RGB camera frames
--compress	none	JPEG quality 1-100

kino ingest

Convert external datasets into .kdb.

kino ingest <SRC> \
  --output output.kdb \
  --format hdf5 \
  --embodiment franka \
  --task "open drawer" \
  --fps 20.0 \
  --max-episodes 100 \
  --compress 85

Option	Default	Description
SRC	required	Source file or directory
-o, --output	output.kdb	Output .kdb path
-F, --format	hdf5	hdf5, lerobot, rlds, or tfrecord
-e, --embodiment	unknown	Robot embodiment
-t, --task	inferred when possible	Task description
--fps	10.0	Control frequency
--max-episodes	none	Limit ingest
--compress	none	JPEG compression quality

kino info

Print a database summary.

kino info data.kdb
kino info data.kdb --episodes

Option	Description
--episodes	Include per-episode table

kino schema

Print structural information, including file version, index offset, field dimensions, cameras, and byte budget.

kino schema data.kdb

Use this when you need to answer “what is inside this file?” before training.

kino validate

Check file consistency and decode every episode.

kino validate data.kdb
kino validate data.kdb --verbose

Validation checks include:

• header/index episode count consistency,
• total frame count consistency,
• metadata decode,
• full episode decode,
• action/state dimensions,
• NaN/Inf warnings,
• image shape consistency,
• terminal-frame warning.

kino query

Run a KQL filter.

kino query data.kdb "success = true AND num_frames > 50"
kino query data.kdb "task CONTAINS 'pick'" --limit 20

Option	Description
--limit	Maximum results to print

kino mix

Inspect weighted virtual mixtures.

kino mix --source bridge.kdb:0.4 --source aloha.kdb:0.6
kino mix --source bridge.kdb:0.4 --source aloha.kdb:0.6 --sample 1000

Option	Default	Description
-s, --source	required	path:weight, repeatable
--seed	42	Sampling seed
--sample	none	Sample N episodes and print distribution

kino merge

Write multiple .kdb inputs into one output.

kino merge lift.kdb pusht.kdb --output combined.kdb
kino merge lift.kdb pusht.kdb --output successful.kdb --filter "success = true"

Option	Default	Description
inputs	required	Input .kdb files
-o, --output	merged.kdb	Output path
-F, --filter	none	Optional KQL filter

kino export

Export a .kdb file back to simple interchange formats.

kino export data.kdb --output export/ --format numpy
kino export data.kdb --output metadata/ --format json

Format	Output
numpy	Episode directories with actions.bin, states.bin, rewards.bin, camera .bin files, and JSON metadata
json	Metadata-only JSON

kino bench

Run a synthetic benchmark for write, open, sequential read, random read, metadata read, and KQL.

kino bench -n 500 --frames 50
kino bench -n 100 --frames 20 --images

The benchmark writes to /tmp/kinodb_bench.kdb and removes it at the end.

kino-serve

Start a gRPC server.

kino-serve data.kdb --port 50051

kino-serve \
  --source bridge.kdb:0.4 \
  --source aloha.kdb:0.6 \
  --port 50051

Option	Default	Description
file	none	Single .kdb file
-s, --source	none	Weighted source, repeatable
-p, --port	50051	Listen port
--bind	0.0.0.0	Bind address
--seed	42	Mixture sampling seed