← Back to changelog
v1.16

Speech Plugins, Time Preferences, & Capture Privacy

May 22, 2026

v1.16 splits speech into dedicated plugins, adds user timezone and clock preferences, makes screenshots private by default, improves model and plugin controls, and clarifies how Editor and Office artifacts are routed.

🎙️ Speech Plugins

  • Speech split into built-in plugins — The legacy core speech system is now separated into _kokoro_tts for Kokoro TTS and _whisper_stt for Whisper STT.
  • Independent TTS and STT toggles — Kokoro TTS and Whisper STT can be enabled separately. Browser-native TTS remains available as a fallback when Kokoro is off, and the Whisper microphone UI appears only when the plugin is enabled.
  • Engine-agnostic speech services — Legacy speech settings and APIs were removed in favor of service brokers that are not tied to one engine.

🕒 Time & Preferences

  • User timezone preference — Scheduler views, notifications, backups, memory timestamps, and frontend time displays now render user-facing dates in the selected timezone.
  • 12/24-hour clock preference — Time displays respect the user's chosen clock format while UTC remains the internal boundary for absolute instants.

🛡️ Capture Privacy

  • Ephemeral screenshots by default — Browser, Desktop, and computer-use screenshots now stay context-scoped by default through in-process image references consumed by vision payloads instead of being written to disk.
  • Durable screenshots remain explicit — User-initiated screenshot paths still produce durable files when that is the requested action.

🧩 Models & Plugins

  • Model preset deep merging — Preset switches now preserve custom context windows, rate limits, and nested keyword arguments across model slots.
  • Inherited utility and embedding slots — Omitted utility and embedding slots inherit from the currently configured models instead of being overwritten.
  • Plugin scanner calibration — Scanner prompts focus on demonstrated risk rather than flagging common capabilities such as network calls or filesystem access by default.
  • Simpler plugin activation — Plugin lists now use a one-click ON/OFF switch instead of the previous dropdown control.
  • Configurable error retry limit — The _error_retry plugin reads its retry limit from settings instead of using a hardcoded single retry.
  • Skill visibility controls — Skills hidden from the chat Skills selector are filtered from model-facing listing, search, recall, and prompt injection.
  • Connector skill activation — Connector clients can activate skills in live chats through the new skills_activate endpoint.

📝 Editor & Artifacts

  • Text Editor owns Markdown saves — The Editor now owns Markdown save flow, canvas-open intent, refresh behavior, and stale-save protection.
  • Office artifacts renamed — Office artifacts are now office_artifact and remain Desktop-only for Office formats.
  • File Browser routing clarified — Markdown opens in Editor, text and Office documents open in Desktop, and browser-renderable files open in Browser.
  • Root Markdown files open in Editor — Root-level Markdown files can now be opened from the File Browser into the Editor surface.

🎨 UI & Runtime

  • Unified surface headers — Browser, Desktop, and Editor floating modals now share a consistent header action rail with aligned ordering and separators.
  • Composer progress markers improved — Icon markers now render as Material Symbols in composer progress overlays.
  • Browser screenshot behavior refined — Browser screenshots open in the image viewer, while live previews continue to open the Browser surface.
  • Codex OAuth setup expanded — The setup UI adds Main and Utility model selectors plus account quota usage bars on the dashboard.
  • Host browser recovery clarified — Browser settings now give clearer recovery guidance when host-browser runtime startup fails and direct users back to the internal Docker browser.

🐛 Fixes

  • Message queue persistence fixed — Queue state now persists correctly after auto-send.
  • WhatsApp sender number included — WhatsApp connector user messages now include the sender number.
  • Computer Use rearm guidance updated — Guidance now points users to the /computer-use on CLI access model.

⚠️ Breaking Changes

  • Legacy speech settings and APIs removed — TTS and STT are managed through the _kokoro_tts and _whisper_stt built-in plugins. Existing speech configuration will not carry over automatically.
  • document_artifact renamed to office_artifact — Retired shims and facades were removed. Skills and integrations using the old name need to update.