Manual[01]
Finding a shot means scrubbing
The shot you remember is somewhere in four hundred clips. Finding it means scrubbing, even on the trip you shot a year ago.
Prewatch · Indexed
A014_C006 · 4K DCI · 2.39:1 · 25p
Steady · Talking to camera · English
40mm · ISO 800 · 1/50 · f/5.6
Move across the picture to scrubDrag across the picture to scrub
A014_C006 · 00:04:00:00
4096 × 1716 · 2.39:1
00:00:00:00
Works withConnect once
ClaudeClaude Desktop and Claude Code
ChatGPTThe ChatGPT Mac app and Codex
DaVinci Resolve Studio 21.1Through its own MCP server
Any drive or NASThe Prewatch folder travels with the media
Footage from
- Sony
- Canon
- Panasonic
- Nikon
- Fujifilm
- Blackmagic Design
- ARRI
- RED
- DJI
- GoPro
- Insta360
- iPhone
H.264, HEVC, ProRes and MPEG-2 in MOV and MP4, and AVCHD (.MTS), today. In development: Blackmagic RAW, REDCODE RAW, ARRIRAW, Sony X-OCN, Canon Cinema RAW Light, ProRes RAW, MXF. Final Cut Pro and the free DaVinci Resolve through a file handoff: planned.
Problems
Your edit software doesn't know what's in your footage
An editing app plays whatever you point it at. Finding the shot is still your job: scrubbing, logging and remembering which card had the good take.
Disconnected[02]
Your AI can't see your footage
Claude and ChatGPT can plan a cut, but they have never seen your rushes, so you end up describing clips to them.
Unlogged[03]
Nobody watched the rushes
Without an assistant editor, nothing records who speaks, what they say, or where the moments are on each card.
Unchecked[04]
Bad takes surface late
A soft, shaky or underexposed take turns up once it is already in the timeline.
The film
From a folder to a cut, in two minutes
Watch Prewatch index a shoot and transcribe what was said. You tick the best quotes in the transcript, and your AI, Claude or ChatGPT, finds the moments and builds the rough cut around them in DaVinci Resolve. One take, nothing staged.
Prewatch · product film2:00 · 1080p · captions
The film arrives with the app
- 00:00Drop in a folder
- 00:20Every clip indexed on the Mac
- 00:45Tick the best quotes in the transcript
- 01:10Ask your AI for the moments
- 01:40The cut appears in Resolve
How it works
Index once. Ask anything. Cut in Resolve.
Prewatch doesn't replace your editing app or your judgement. It does the watching before the edit starts, then your AI does the searching and the assembling while you decide.
Index[01]
- Shots and keyframes
- 412
- Transcribed
- 38 clips · 3 languages
- Speakers told apart
- in 38 clips
- Days and scenes
- 6 days · 41 scenes
- Flagged
- 7 soft · 3 dark
Point it at a folder
Every clip is watched on your Mac: shots, speech, speakers, days and scenes, quality. Speech is transcribed in 100 languages, with English translations. Clips you add later are indexed too.
Ask[02]
Where does she mention the boat?
A022_C01401:12–01:19“we nearly missed the boat”A031_C00300:08–00:15“the boat back is at six”Ask in plain words
Your AI searches what was said and seen across every clip, and answers with the clip, the seconds and the timecode.
Cut[03]
00:0000:3001:00
5 clips placed · markers on the 3 moments
Build the cut together
Your assistant lays the clips on a timeline in the Resolve project you have open, through Resolve's own MCP server. You keep editing from there.
Prewatch for Claude and ChatGPT
Find where I talk about the hike on day 3. I want the best take.
prewatch · search "hike" day 3
| type | clip | in | out | q | text |
|---|---|---|---|---|---|
| moment | A014_C006 | 44.0 | 72.0 | 5 | talks to camera, the summit behind |
| said | A014_C006 | 47.2 | 50.1 | - | three days of hiking to see this |
| said | A014_C011 | 12.5 | 19.8 | - | the hike in was brutal, honestly |
prewatch · frame A014_C006 at 52.0
You talk about the hike twice on day 3. The better take is A014_C006 from 44.0 s: steady, in focus, with the summit behind you. I checked the frame. Want it on a timeline, followed by the cloud clearing off the summit?
One click in Prewatch's settings connects it to Claude Desktop or the ChatGPT app. Your assistant gets an overview of the footage, searches what was seen and said, and gets back moments with in and out points, source timecode and the file to import.
Setup guide →
A014_C006 · 9 keyframes
3 of 9 moments
What Prewatch finds
Your AI, with your footage in mind
Nobody has to log anything on set. Prewatch watches the footage you already have, yesterday's cards or last year's trip, so you and your AI talk about the story, not the timecode.
Prewatch runs on your Mac and indexes every clip in your footage folders. Claude or ChatGPT on your Mac connects to it through MCP and gets answers with clips, seconds and timecode, not guesses.
A014_C006 · what Prewatch knows
- Shots and keyframes1 shot · 9 keyframes · contact sheet
- Camera motionSteady
- Quality flagsOverexposed 1:46–1:52
- Shot roleTalking to camera
- SpeechEnglish · 1 speaker
- Said at 0:47“Three days of hiking to see this”
- Moments9 · the best at 0:44 and 2:30
- Camera40mm · ISO 800 · 1/50 · f/5.6
- Day and sceneDay 3 · scene b · 14:32
Every clip in the folder gets the same record, in plain files next to the footage. See what is in the Prewatch folder.
Why Prewatch
Your AI reads a page, not the footage
Watching video is the slow, expensive part. Prewatch does it once, on your Mac, with on-device models for speech, speakers, camera motion and image tags. Your AI then works from compact text instead of video, so even a long shoot fits in one conversation, and no video or audio is uploaded.
For a five-hour shoot, an AI that watches and listens to the footage needs over five million tokens, far more than fits in a conversation. Prewatch's index for the same footage is about 35,000: a short page per folder, one line per clip and one per moment, with the detail opened only for the clips your AI needs.
152×less for your AI to read
- The footage
- A real trip folder: 381 clips, 4 h 58 min of picture and sound.
- Watching it
- Picture and sound at the rates Google publishes for Gemini, a model that takes video directly: 263 tokens a second for the picture, 32 for the sound, so 295 tokens for every second of footage. The Claude and OpenAI APIs don't take video directly; they would need frames and a transcript.
- Prewatch
- What your AI reads to plan a cut: the folder's index, its scenes and one line per clip, counted with Claude's tokenizer, October 2026. Transcripts and moments are searched, not read whole.
Privacy
Indexing stays on your Mac. You choose what AI sees.
Your video and audio never leave the Mac, and nothing is sent to us. If you turn on descriptions, still frames and transcript text go to Claude or OpenAI with your own key, at a cost shown before you turn it on. When you chat, your assistant reads Prewatch's answers and any frame it asks to see.
| Step | Where it runs | What leaves the Mac |
|---|---|---|
| Decoding, shots, motion, flags, tags | On your Mac | Nothing. The video is decoded inside the app with macOS's own frameworks, and image tags come from Apple's on-device Vision. |
| Transcripts and speakers | On your Mac | Nothing. Whisper and the speaker models run on the Mac. Voiceprints, which recognize a voice across clips, stay on this Mac and can be deleted in Settings. |
| Days and scenes | On your Mac | Nothing. Grouped from the clips' own times and locations. |
| Descriptions and moments | Claude or OpenAI | Only if you turn it on: still frames and transcript text, sent with your own key. Never video, audio or locations. |
| Questions and timelines | Your AI app | Prewatch's short text answers, and a frame when the assistant asks for one, become part of your Claude or ChatGPT conversation. |
Price and requirements
One app, two ways to pay
Every feature on both plans, starting with 14 days free. Apple handles the subscription; there's no account with us and no credits.
YearlyBest value
$59 / year
$4.92 a month, billed once a year.
Monthly
$7.99 / month
For one project or a busy season.
What you need
- A Mac with Apple silicon and macOS 26 or later
- Claude Desktop or the ChatGPT app, on the plan you already have
- DaVinci Resolve Studio 21.1 or later to build timelines; the free version still indexes, searches and asks
- Optional: your own Claude or OpenAI API key for descriptions and moments, billed by them, about a cent for a long take
14 days free, then $59 a year or $7.99 a month. Works with your Claude or ChatGPT plan; timelines need DaVinci Resolve Studio 21.1+. Clip descriptions are optional and use your own API key, billed by Anthropic or OpenAI.
Who it's for
Made for the people in the edit
- Documentary editorsSearch weeks of interviews for any name, place or line, and get the clip and the seconds to cut on.→
- Travel and adventure filmmakersDays and scenes come from the clips' own times and places, so “day 3 at the lake” means something.→
- Wedding and event filmmakersVows and speeches transcribed, speakers told apart, so the line that got the laugh is one question away.→
- YouTubers and creatorsTurn a card dump into a rough cut in one conversation, then keep editing in Resolve.→
- Assistant editorsTranscripts, flags and selects written before the edit starts, in plain files the whole team can open.→
- Small studiosIndex once on one Mac. The Prewatch folder travels with the media, so every Mac reads the same index.→
What does Prewatch do?
Prewatch is a Mac app that watches your footage before you edit. For every clip it finds the shots, keyframes and camera motion, flags soft, dark or frozen footage, reads the camera metadata, transcribes speech, tells speakers apart and groups the clips into days and scenes. With your own Claude or OpenAI key it also describes each clip and marks the moments worth cutting, with in and out points. Claude or ChatGPT then answers questions about your footage from this index and, in DaVinci Resolve Studio, builds the cut with you.
Do I need to mark or log anything while shooting?
No. Nobody has to tap, mark or take notes on set. Prewatch works on the footage you already have: today's cards, last year's trip, a shoot nobody logged. Point it at the folder and it watches every clip itself.
Is my footage uploaded anywhere?
No video and no audio ever leaves your Mac. If you turn descriptions on and add your key, Prewatch sends still keyframes, one overview contact sheet and the transcript text of each clip to the provider you chose, Claude or OpenAI. Locations and anything about speakers are never sent. When you chat in Claude or ChatGPT, the assistant reads what Prewatch answers, as part of your conversation. Every step and where it runs.
Do I need a Claude or ChatGPT account, or an API key?
To chat with your footage you need Claude Desktop or the ChatGPT app for Mac, signed in to your account. Prewatch connects to either with one click. Claude Code and the Codex CLI work too. ChatGPT in a web browser can't reach apps on your Mac.
An API key is optional. Indexing, transcripts, speakers, scenes and search all run on your Mac without one. Descriptions, moments and the better English translations use your own API key from Anthropic (Claude) or OpenAI; that company bills you for the use directly, and Settings shows what a clip costs before you turn it on.
Does it work with the free version of DaVinci Resolve?
Building timelines by chat needs DaVinci Resolve Studio 21.1 or later, whose built-in MCP server lets your assistant work in Resolve. With the free version you can still index your footage, search it and ask Claude or ChatGPT about it, then place the clips yourself. A file handoff that the free version can import is planned.
Does it work with Premiere Pro or Final Cut Pro?
Not directly. The chat-to-timeline workflow is built for DaVinci Resolve Studio. A file handoff that Final Cut Pro and the free Resolve can import is planned; there is no Premiere Pro integration. Whatever you edit in, the index is plain files, so you can search your footage, find the clips and timecodes and bring them in yourself.
How does the free trial work?
Prewatch is an App Store subscription with a 14-day free trial: $59 a year or $7.99 a month after it. You manage or cancel it in your Apple Account's subscription settings; cancel before the trial ends and you are not charged.
Prewatch watches everything.
Your AI builds the edit with you.
14 days free, then $59 a year or $7.99 a month. Works with your Claude or ChatGPT plan; timelines need DaVinci Resolve Studio 21.1+. Clip descriptions are optional and use your own API key, billed by Anthropic or OpenAI.