1.7 KiB
@kitlangton/hex
0.3.1
Patch Changes
04a42da: Decode model progress and dictation levels with shared bounded SSE framing. Large transport chunks containing many small level events no longer fail the event limit, and CR-delimited events dispatch without waiting for more data or EOF. Cancellation, callback failures, and bounded observation buffers remain supported.
0.3.0
Minor Changes
-
c5bffe3: Add a model-boundcreate({ model, language, onProgress })API that returns a ready transcriber for WAV bytes. The Promise transcriber provides explicit, idempotent cleanup; the Effect transcriber is scope-owned. Failed creation and in-flight transcription cancellation wait for helper cleanup. Existing low-levelcreate()and desktopconnect()remain available.Effect scope cleanup failures now surface as defects instead of being logged and discarded. The low-level and model-bound option types are distinct so shared configuration cannot accidentally change the returned resource's shape.
Startup cleanup failures are surfaced instead of hiding a potentially live child. Cancellation also takes precedence over a buffered result that has not settled.
Embedded use still requires an explicit native helper command until bundled platform artifacts are published.
0.2.2
Patch Changes
e163753: Support Effect 4.0.0-rc.112 and newer in the Effect client, using the current tagged-error API.
0.2.1
Patch Changes
c5ba7d4: Stop dictation heartbeat timers after terminal capture errors while preserving retries for transient failures.
0.2.0
Minor Changes
- Require local API v2 for owner-scoped dictation capture, rejecting legacy HEX apps before a recording can start and become orphaned.