Shipped, constantly.

1.1.1

One release for every edition: the direct download, the FxFactory edition and Studio are now the same app, and Studio adds Share. Read the full 1.1.1 release notes →

• Use Doza Assist in ChatGPT and Claude. One click connects Doza Assist to ChatGPT for Mac, alongside the Claude connector: work with your interviews in the frontier AI you already use. Only the projects you switch on are visible, and your video never leaves your Mac.
• 25 languages get blazing fast transcription. Parakeet v3 brings the fast engine to French, German, Spanish, Italian, Portuguese, Dutch, Polish, Ukrainian and 16 more European languages. Before, only English ran on Parakeet.
• “What is this piece for?” A short prompt at the top of every project tells Doza Assist what you are making, and AI Analysis, Story Builder, the Story Brief and chat gear their picks toward that goal.
• Story Builder sees the whole interview. The app maps every minute of an interview into moments, so a story can draw on any part of it. Stories follow the subject you ask for and land within 10% of the requested length; extend a story instead of starting over.
• Clips play exactly as cut. Clips start on the first sound of the first word and stop within a few milliseconds of their out point, everywhere in the app. Every AI clip contains the words it quotes.
• Speakers get real names, including yours. The interviewer named in the project goes to the person asking the questions; Find speaker names asks again with one click.
• My Style follows your own notes first and sizes clips to your typical clip length.
• Stable on 8 GB Macs, and self-repairing. A full analysis runs cleanly on 8 GB Macs. If the AI or speech model’s files go missing, the app downloads only what is missing at the next start.
• Your own key: Claude Opus 5.5 / Sonnet 5.5 or GPT-6, with prompt caching, and a sturdier assistant connector.
• Import and export: synchronized clips import straight from a Final Cut Pro browser export; exports keep clip titles and never drop a select an assistant just added; a collection story exports as one timeline in story order.
• Studio: everything above, plus Share updates: a Documents tab on share links, collections that share fully, Teams keys with seats, and a free trial with two share links.

Pricing

• Doza Assist is $199, one-time. The launch discount code has ended; checkout opens at the list price with nothing to apply.
• Studio is $299 a year or $29 a month. Pick monthly or annual on the pricing page. Anyone who subscribed at the launch price keeps it.
• Updates come with Studio. The $49 a year updates renewal now applies only to a perpetual license with no active Studio subscription.

Studio and Share

Three plans, one app. Read the announcement →

• Doza Assist (Studio). A subscription edition of the same app with hosted Share built in: publish a link, collaborators pull selects, comment, download documents and connect their own AI assistant in a browser, and everything syncs back to your project. Every update is included while you subscribe; one Mac at a time.
• Share. Compressed audio, transcript, analysis clips and documents go to share.doza.ai; the video never leaves your Mac. Links are revocable, purge on unpublish, and carry a Documents tab (transcript, selects and quote sheet as Word, PDF and Excel).
• Try Studio free. The Studio trial includes two share links with every Share feature on the first two minutes of any file. Download the Studio build.
• Enterprise. One key for a whole team, seats per machine, invoices and purchase orders welcome; by email.
• Licensing. Perpetual keys keep working forever; Studio and Enterprise keys carry a 14-day grace window after a missed payment; the license server handles renewals, refunds and seat counts on its own.

1.1

The biggest update since launch. Four things headline it: a much smarter Chat, Collections that work like a project, a connector that lets Claude read the projects you choose, and transcript correction by double-clicking a word. Read the full 1.1 release notes →

• Chat reads the whole interview, however long. One box, five story-first questions, short answers in plain words with playable clip cards; follow-up questions under every answer; it knows when not to give clips, remembers what you are making, and lets you pass on moments.
• Correct the transcript in place. Double-click any word to fix it, insert a word the engine missed, split a paragraph, or give a line to a different speaker (including a new one). Cmd+Z undoes each step. Media timing never changes, so every clip keeps its timecode; a line above the transcript offers to re-run the analysis.
• Collections work like a project. Check two or more interviews on the Projects page and Create Collection. Same tabs, same chat, same Story Brief, Story Builder, Clips and Export, all working across every interview at once.
• An AI assistant connector (MCP). Switch a project on and Claude Desktop, Claude Code, Codex, Cursor and VS Code can read its transcript and selects, search it, and add clearly labeled selects to your Clips. One-click add to Claude Desktop. Off by default, per project; media files never leave your Mac.
• Exports named the way you look for them. “Project – Selects 1”, “Project – Markers 1”, “Project – Story: Title”, numbered per project; every exported clip carries its words in the note; Story Builder sequences export from the Export tab; press-format quote sheets.
• Clips read like clips. Generated titles, the first line, Show text everywhere; a word-level Trim sheet; sub-second in and out points; drag to reorder.
• Speakers named from the transcript. After speaker identification the app names the speakers it can from introductions and direct address, never inventing one; hand-set speakers survive a re-run.
• Story Brief names each moment’s real speaker, opens a word-for-word transcript, adds single moments to Clips, and never replaces a good brief with an empty one.
• Transcription: Auto-detect language listens before it picks an engine; 36-hour files transcribe; guardrails warn about very long projects and unsynced FCPXML media. AI Analysis is faster on long interviews.
• Interface: tabs in the order of work, a gear menu that shows current values with Project details and Relink media, dark by default, honest first-run setup.
• Fixes: MPEG-2 camera files (Sony XDCAM MXF, MPEG-TS) report their frame rate correctly; long-interview analysis keeps everything the model finished; the compressed-window player layout; retranscribe no longer strands the analysis; folder drag fixes; “Send to Final Cut” opens the copy you use; and more in the full notes.

1.0.46

• Your work now lives in ~/Documents/Doza Assist: projects and every exported file (timelines, transcript documents, quote sheets) moved out of the hidden app-support folder into a normal Documents folder — visible in Finder, included in Time Machine, and synced by iCloud Drive if your Documents folder is. Existing projects migrate automatically on first launch; macOS asks once for Documents access, and declining keeps everything working from the old location. No relinking — your source footage never moves.
• Import multicam clips straight from the Final Cut browser. Select one or more multicam clips, choose File › Export XML, and drop the file in — no scratch timeline needed. An export holding several multicams becomes one project per clip, named after the clip.
• Collection exports now rebuild multicam timelines properly: interviews imported as multicams come back to Final Cut with video and audio angles connected, instead of audio-only files.
• Clip exports match your selection exactly. The Clips tab’s Export button exports the checked clips directly (with their colors and titles), sub-second in/out points are preserved, and your personal labels no longer ride along uninvited.
• “Send to FCP” now finds every Final Cut edition, including the Apple Creator Studio subscription version and the trial.
• Creating a collection now warns before you navigate away mid-build, and finished work resumes with one click instead of looking lost.
• Reliability: closing the window during an engine restart no longer leaves the app windowless, and the User Guide was updated for everything above.

1.0.45

• Fixed a blank window when reopening the app: closing the window and clicking the Dock icon now restarts the analysis engine behind a brief “Restarting…” screen instead of showing an empty dark window. Closing the window now also fully frees memory — the AI model no longer stays loaded in the background.
• Exports follow your “Edit in” choice everywhere. With DaVinci Resolve selected, FCPXML round-trip exports are saved and revealed in Finder with Resolve focused and clear import guidance — the export button now says exactly what will happen instead of always sending to Final Cut Pro.
• Documents → Selects now works on projects imported from an FCPXML timeline (it previously reported “check at least one category” even with clips selected) and includes the Soundbites category.

1.0.44

• Free trial: Doza Assist can now be downloaded and tried before buying. Every feature works on the first 2 minutes of each file; entering a license key unlocks full-length transcripts in place — no re-download.
• Multicam angles built from many stop-start camera files now transcribe in full. Previously only the first file of an angle was picked up, so long interviews came through truncated.
• Fixed transcript timecodes drifting on long audio files (MP3): clicking a passage now plays exactly the words on screen, even hours in.
• New Documents export on the Export tab: the full transcript, your selects, or the story quote sheet as Word, PDF, or Excel — saved to your Exports folder and revealed in Finder. PDFs now render non-Latin languages correctly.
• Transcript downloads (TXT/SRT/JSON) always use your current speaker names, and JSON includes the resolved names per segment.
• Broader timeline compatibility: FCPXML with spine-level clips (including Doza's own flat exports), projects starting at 01:00:00:00, and older DaVinci Resolve XML now import correctly.

1.0.43

• Smarter model loading on 16GB Macs: the boot-time AI warm-load now runs only when there's comfortable memory headroom, keeping the rest of your system responsive on smaller machines.
• Tighter bounds on background analysis memory use, plus new memory telemetry in the logs to make support faster.

1.0.42

• Launch reliability: opening the app while another copy is starting now hands off cleanly to the running instance in every case.
• Update plumbing hardening so future updates install even more dependably.

1.0.40

• New memory governor for 8GB and 16GB Macs: Doza now detects your machine's memory tier and gates transcription, diarization, and analysis so big jobs finish without starving the system.
• Validated end-to-end on real 8GB and 16GB hardware — zero Metal out-of-memory events across the full pipeline.

1.0.38

• Open-source compliance release: expanded NOTICES and complete third-party license texts now ship inside the app, covering every bundled component.

1.0.37

• Storytelling engine v1.1: the guidance behind Story Brief and story analysis was rewritten top to bottom for sharper beat detection and cleaner select rationale.

1.0.35 – 1.0.36

• AI runtime upgrade: models now load without re-downloading after updates, and prompt caching keeps follow-up questions fast.

1.0.32 – 1.0.33

• Chat speed overhaul: the first answer in a project now arrives in seconds instead of half a minute, and responses stream in live as they're written.
• Grounded chat rework: answers stay anchored to your actual footage, with clip references preserved end-to-end — vague summaries replaced by quotes with timecodes.

1.0.28 – 1.0.29

• Duration targets: give Story Builder or chat a runtime ("cut this to 90 seconds") and the target is enforced in the output, not just suggested.
• Clips tab: drag-to-reorder your selects directly.

1.0.27

• Export hardening: 85 verified fixes across FCPXML and Premiere export — boundary-accurate clip splitting, probe-based format declarations, and per-NLE gating so each editor gets exactly the file it expects.

1.0.25 – 1.0.26

• Transcription reliability: fixed silent-proxy audio selection (originals are now always preferred), added guards against hallucinated text on silent footage, and fixed speaker diarization on Macs without Homebrew installed.

1.0.23

• FCPXML round-trip fixes for rate-conformed clips, and support for browsing .fcpxmld bundles directly.

1.0.19

• DaVinci Resolve: version selection and export fixes. Plus an audio-track picker for multi-track sources, MXF video preview, audio-first playback, and correct start-timecode handling for MXF media.

1.0.17 – 1.0.18

• Output Language: set any of 30 languages per project and every AI deliverable — summaries, ranked soundbites, Story Brief, chat — comes back in that language. Transcription itself understands 100+ languages, auto-detected.