LinkSense v0.3.8

Highlights

Normalize supported voice recordings to 16 kHz mono PCM WAV before transcription, with bounded conversion time, duration validation, and temporary-file cleanup. The API image includes FFmpeg.
Support OpenAI-compatible speech recognition gateways that return a standard JSON transcription response, while retaining native DashScope support.
Add an OpenAI-compatible image provider and configurable service addresses for image providers, including existing encrypted configurations.
Allow reasoning models to complete automatic task titles without the previous fixed output-token limit.
Update image-provider configuration guidance and translations in all six supported locales.

CI and source security checks passed for the released commit.
Linux amd64 and arm64 Core/Full installation, repair, and upgrades from v0.3.7 passed, including persisted-data verification.
Release files are checksummed and use immutable image digests. Docling remains at 1.36.0.
Under the existing application-owned image policy, known external-image and verified inherited-baseline vulnerabilities remain documented and report-only; they have not been remediated by this release. Application-introduced HIGH/CRITICAL findings, scan failures, and functional failures remain blocking.

Full Changelog: v0.3.7...v0.3.8