0.60.0

Breaking Changes

• The session.created, session.updated, and session.file_added platform events are removed, along with OncePer.Session and the dry-run session_id option. A trigger subscribed to one of them no longer parses.
• Sessions (experimental): SessionFile.range_min_timestamp_ns and range_max_timestamp_ns are replaced by min_file_timestamp_ns and max_file_timestamp_ns, which are in the file's own timestamps, not Unix-epoch nanoseconds. Renaming the fields without converting the values declares the wrong window.
• Sessions (experimental): Session.add_file, add_files, remove_file, and remove_files no longer return the Session. The plural forms return a BatchResponse and do not raise when the platform refuses an entry, so check its failed list.
• Device.create_session (experimental) requires name, and a second call with the same name returns the existing session instead of creating another.
• File.set_timeline_offset (experimental) takes offset in place of unix_epoch_offset_ns, and reads a float, Decimal, or numeric string as seconds, where it read nanoseconds.
• Metric.publish and Session.publish_metrics (experimental): each item in the result's failed list is the exception the platform refused it with, replacing PublishMetricsError.
• Topic.from_id, Topic.from_record, and the Topic constructor (experimental) take context in place of session_context. Topic.context and Topic.set_context are removed; build another Topic to read in a different context.
• Topic.get_data, get_data_as_df, and get_data_as_record_batches (experimental) return only the fields that fields_include and fields_exclude select. The field a topic's timestamps come from is no longer added to a filtered read, and filters that select no field raise RobotoInvalidRequestException.

Features Added

• Device.create_sessions (experimental) creates up to 100 sessions in one call from files you have already uploaded, declaring each file's topics, schemas, and timeline sources. File.declare_topics declares one file's topics without a session, and Session.add_files accepts the same declarations.
• A declared topic (experimental) can list representations, the MCAP or Parquet files a read opens, so data in a ROS bag, PX4 ULog, or CSV is read through files converted from it. File.set_representations replaces them later.
• Session.set_unix_offset and Topic.set_unix_offset (experimental) anchor data recorded in relative time to wall-clock time; anchor_ns on a declaration does the same when the data is declared.
• Device.get_or_create (experimental) registers a device, or returns the existing one when the ID is taken.
• Every method taking a roboto.time.Time accepts a NumPy integer, so a timestamp from a pandas or NumPy column can be passed as is.
• FileContext, DatasetContext, and DeviceContext (experimental) scope a Topic's reads to its data in one file, a dataset's files, or a device's files, with no session needed. start_time and end_time may be omitted and default to the extent of that data.
• roboto.ai.mcp gains McpServerOrgCredentialRecord and SetMcpServerOrgCredentialRequest, record types for an MCP server's org token: an org secret whose value the AI sends as a bearer token for service users, and for members who have not connected the server themselves or whose own connection fails. The record names the secret and never holds its value. These are request and response types only: no SDK method sends or returns them.

Bugs Fixed

• Experimental APIs no longer emit ExperimentalWarning when called, which could print on every call. Each one is still marked as experimental in its documentation, and roboto.ExperimentalWarning is still importable.
• Experimental topic reads scoped to a session return only that session's data when the session holds part of a file, whether a time window or a data_range slice. Requires this SDK version.
• Experimental topic reads no longer raise TypeError on Python 3.10 and 3.11 when timestamps come from a schema field that declares a time unit.