Event streams | Docs

For reading time series at volume, such as joint states, IMU, poses and events, use the Arrow Flight endpoint instead of JSON. Data arrives as Arrow record batches that load directly into PyArrow, Polars or pandas without parsing.

Endpoint

FLIGHTgrpc+tls://flight.rayform.world:443

Authenticate with your API key as a bearer token in the call headers.

Tickets

A ticket is a JSON object naming what to stream:

Field	Type	Required	Description
`segment`	string	yes	Segment ID
`version`	string	no	Segment version; defaults to the latest
`channels`	array	yes	Channels to stream, or `["events"]` for interaction events
`t0_ns`, `t1_ns`	integer	no	Time window; defaults to the whole segment
`frame`	string	no	Output convention, as in export options

Reading with PyArrow

Python

import json
import os

import pyarrow.flight as flight

client = flight.FlightClient("grpc+tls://flight.rayform.world:443")
options = flight.FlightCallOptions(headers=[(b"authorization", f"Bearer {os.environ['RAYFORM_API_KEY']}".encode())])

ticket = flight.Ticket(json.dumps({"segment": "seg_4f0a", "channels": ["joints/position", "joints/velocity"]}).encode())
table = client.do_get(ticket, options).read_all()
print(table.schema)

Text

t_ns: int64
joints/position: fixed_size_list<item: float>[7]
joints/velocity: fixed_size_list<item: float>[7]

Reading with the SDK

The SDK wraps the same calls:

Python

seg = client.segments.get("seg_4f0a")
joints = seg.channel("joints/position")          # pyarrow.Table
events = seg.events()                             # pyarrow.Table, time-ordered

Event schema

Column	Type	Description
`t_ns`	int64	Event time on the session clock
`type`	string	Event type, for example `grasp`
`entities`	list of string	Entity IDs involved
`t_start_ns`, `t_end_ns`	int64	Span of the interaction
`confidence`	float32	Detector confidence, from 0 to 1