Ingestion
Data is consumed progressively without requiring the complete stream to be available in memory.
Engine / library
An independent boundary-driven continuous-stream segmentation engine, designed to consume data incrementally and produce bounded segments without depending on source-specific semantics.
Brynse incrementally consumes encoded streams, retains only the data required for analysis, resolves pluggable boundary signals and materializes bounded segments. The engine remains generic while integrations provide domain-specific semantics.
Data is consumed progressively without requiring the complete stream to be available in memory.
A bounded ring retains working analysis data while reclaimable encoded content can be temporarily spooled to disk for segment materialization.
Fixed, manual and external boundary providers generate candidates that are resolved through evidence and segmentation policy.
Once a boundary is resolved, Brynse plans byte ranges and materializes final segments aligned with the encoded format.
The first-party radio integration adds domain-specific semantics on top of the generic engine without changing its segmentation model.
Incremental ICY metadata parsing and semantic tracking of track transitions.
Candidate selection using acoustic signals and bounded analysis windows.
A router classifies transitions and delegates temporal geometry to dedicated specialists.
Hard cut, fade, interstitial and bounded insert specialists handle different transition phenomena.
Ambiguous transitions use a conservative fallback rather than forcing an acoustic classification.
Finalized tracks are written incrementally and persisted as transitions become settled.
Brynse exposes a structured API for embedding the engine in applications, automation and agent-oriented workflows. SegmentRequest defines the request contract while safe entry points return either results or structured errors.