Doza Assist is a local-first AI editing assistant built by a 15-year documentary editor for editors who cannot legally upload footage to cloud transcription services. Everything runs on the user machine — transcription via NVIDIA Parakeet (primary) and OpenAI Whisper (fallback), AI analysis via Ollama serving Gemma 4, audio extraction via ffmpeg, and NLE export via FCPXML to Final Cut Pro, Premiere and DaVinci Resolve. An optional cloud key supports both Anthropic and OpenAI, routing each task to the appropriate model tier, but everything runs local by default.
The Editorial DNA engine learns how an individual editor cuts from their finished projects and applies that pattern to new footage selects, with multiple profiles per editor. The vendor states three finished pieces is where output visibly shifts and five is the recommended baseline — train on one project and results will still feel generic. FCPXML deep ingestion with multicam and sync-clip support means a 4KB XML file replaces what would otherwise be an 80GB ProRes export. Client review runs via Cloudflare Tunnel, self-hosted from the user machine.
Two versions exist. The open source build remains free under MIT with manual Python and Ollama setup. Doza Assist Pro shipped in August 2026 as a signed Mac app at $199 one-time — the licence is permanent, updates are included for the first year, then $49/yr only if you want to keep receiving them. No subscription and no per-minute metering. Install is drag to Applications, with first run handling dependencies and the local model download in roughly 10 to 20 minutes. Sold direct and through FxFactory, which offers a free trial with a capped transcription preview per file.
Pro adds Collections (analysis across a whole folder of interviews), Speaker ID with automatic diarization, Story Brief (recommended arcs, tensions and coverage gaps), Quote Sheets exported to PDF or Word, and selects-aware chat that knows which clips have already been pulled. Both versions gained Output Language across 30 languages with verbatim quotes preserved in the source language, and broadcast media handling including MPEG-TS and MXF with AVC-Intra and embedded timecode.
Requires Apple Silicon and macOS 14 or later, 16GB RAM minimum with 32GB recommended.