Release v0.1.19

Release v0.1.19

• Text-to-speech via any OpenAI-compatible TTS endpoint (Settings → TTS API), exposed to the agent as the tf-assistant-tts MCP tool group (tts_submit / tts_poll / tts_get_result); legacy local TTS worker removed
• Revideo: per-scene timed narration is now mixed into the rendered video output
• Select Text: fixed two crash paths in the selection readout

Release v0.1.18

Hotfix release

• Fix YouTube download failing with opaque 'n challenge solving failed' error when the configured Node.js is older than v22: the app now validates the Node version up front and reports a clear, actionable message
• Trackpad pinch zoom for screenshot pin windows and quick translate

Release v0.1.17

Release v0.1.17

• Revideo: code-first video authoring — write scenes as TypeScript/JSX and let the agent create, edit, and render them via MCP tools (multi-scene timeline, live preview playback, presentation mode, single-frame capture, async MP4/WebM render jobs)
• New Screen & Desktop MCP group (tf-assistant-screen): desktop capture (display / window / region) plus synthetic input (click / hover / scroll / type / press-key) exposed to external MCP clients, gated by an input safety switch
• Agent: user messages now render as Markdown (inline math), the composer slash menu exposes /compact, and a reload-skills action rebuilds the opencode instance
• Fixes: media server answers past-EOF Range requests with 416 instead of aborting; sidecar pins OPENCODE_VERSION when building the opencode fork; screen.input_enabled is now a typed bool so the HTTP settings tab no longer crashes on a stale value

Release v0.1.16

• Inline PDF preview in the file editor: .pdf files now open as a read-only viewer (zoom, page navigation, rotate, two-page spread, text search, thumbnails, outline) instead of 'Binary file (not editable)'
• App-owned MCP tool groups: the app exposes its OCR and PDF parsing as MCP tools that external MCP clients can call, with structured output for the DotsMOCR / PP-OCR engines
• LaTeX math rendering in the file editor and agent chat
• Live PPTX editing: unlock a deck (hand it back to the user) without closing its viewer window
• Agent file browser: 'Open with System' context action, drag-resizable sidebar width, and complete right-sidebar translations across all 7 languages

Release v0.1.14

Release v0.1.14

• Fix registration / permanent activation failing with "no HWID was found on system" on macOS: the hardware ID is now read natively from IOKit instead of shelling out to the ioreg command, so it no longer depends on the process PATH.
• Falls back to an absolute-path /usr/sbin/ioreg (and finally to the previous behaviour) if the native read is unavailable.
• No other user-facing changes.

Release v0.1.13

Release v0.1.13

• Agent auto-continue: when a turn stops mid-task (e.g. it ends on a reasoning message or an incomplete answer), the agent is automatically nudged to keep working until the mission is done
• The loop ends cleanly when the agent returns MISSION COMPLETE, or when a configurable max-nudge cap is reached
• New toggle and max-nudge setting in Agent settings (persist across launches; on by default with a cap of 8)
• Nudge turns render as a distinct dimmed "auto-continue" chip in the transcript

Release v0.1.15

App-managed Agent Providers

• Settings → Agent Providers: manage OpenAI-compatible providers with per-provider model tables, an allowlist mode, default-model selection, and a live connectivity test

Agent auto-continue

• A new session's first turn now resumes auto-continue when it stops mid-task

Release v0.1.12

Release v0.1.12

• Live PPTX editing: open PowerPoint files in a dedicated viewer window and let the agent edit slides visually (add/restyle text and shapes, create charts, apply templates), with host-side fonts and a localized in-viewer UI.
• Attach PDFs in the agent composer: a background-parsed PDF is attached as inline context (or an on-demand file) for the agent.
• The agent keeps one file editor window per workspace, and terminals persist across session switches.
• New MCP tools sidebar in the agent, plus a renamed app MCP group.
• The About settings tab now shows the app version.

Release v0.1.11

New: an in-place screenshot annotation editor.

• Capture a region, then draw rectangles, ellipses, lines, arrows, text, and freehand/polygon ink directly on it; pick any colour; solid, dashed, or dotted line styles.
• Arrows are true block shapes: fill or outline-only, adjustable body thickness and head length, and straight / right-angle / freeform routing with draggable waypoints.
• Rotate any annotation with a handle (angle snapping), or re-style a selected one through a floating, draggable toolbar that follows the system theme.
• The native windows (screenshot, caption overlay, dictation, settings, model downloads) are now localized into seven languages.

Release v0.1.10

New Select Text feature — select any text in any app to summon a floating toolbar.

• Copy the selection, search it on the web, or translate it locally with HY-MT2 (same backend as Snap Translate).
• Send the selection to the AI agent (wrapped in a code block) for follow-up questions.
• The answer panel auto-sizes to its content; the bar follows the system light/dark theme.
• Toggleable from Settings (Select Clip tab); dismisses on scroll or key press.

Also: fix chat deletion after removing its last session, limit the OCR e2e variant picker to downloaded models, and restore focus to the previous app after a screenshot overlay closes.