Privacy policy

The short version

Basecut can analyze your footage in three ways, and what leaves your computer depends on which one you use. Your original video files are never uploaded by any of them, and we never train AI on your footage.

Whichever engine you pick, the app also talks to the internet for things that are not footage analysis: signing in, registering your computer, checking for updates, downloading models, and sending diagnostics (anonymous unless you are signed in, and you can turn them off). Those are listed in Traffic every engine shares. None of it contains your footage. This website also uses standard analytics, described further down.

What stays on your computer

• All source footage you ingest. Basecut never moves or changes it.
• Generated metadata: scene cuts, tags, transcripts, embeddings, ratings, marker positions, color sidecars.
• Proxy and preview files Basecut creates during analysis.
• The Basecut library database (SQLite) that powers the search index.
• Speech transcription (Whisper). It runs on your computer for every engine. The first time it is used, the app downloads the speech model from Hugging Face (see below).

By default these live under your home directory. You can point Basecut at an external SSD in Settings; nothing else changes.

Which engine runs

Basecut picks the engine from your settings. If you have saved your own Google Gemini key and switched it on, your key is used. Otherwise, if Basecut Cloud is your chosen cloud engine and you are on a paid plan (monthly, annual, or lifetime), Basecut Cloud is used. Otherwise Argus is used, or analysis waits until you install Argus or add a key. When Basecut Cloud's monthly allowance runs out, the import carries on with Argus on your computer if Argus is installed, and otherwise pauses with nothing deleted or marked failed.

Engine 1: Argus (on your computer)

Argus is Basecut's own AI model. It runs entirely on your computer, using a local engine that only accepts connections from your own machine. For each section of a clip it reads a handful of still frames, the first 30 seconds of audio and a short block of technical details, all of which stay on your computer. Normal Argus analysis sends no footage, frames, audio, or transcripts anywhere.

That does not mean the app is silent. Argus needs a one-time setup, and the shared traffic below (account, update, and diagnostic requests) still happens while you use it:

• Model download. Argus is about 6 GB. You must be signed in to download it. The app asks Basecut's server (hosted on Supabase) for permission, which checks your plan and device seat and limits how often you can download. The server returns a short-lived link, and the files then come from Hugging Face's download network, so Hugging Face sees that request. Basecut records the file name, size, and outcome of each download as a diagnostic event (see App diagnostics).
• Supporting software. The app may download the local engine software from GitHub and the speech model from Hugging Face.

Engine 2: Basecut Cloud (through Basecut's server to Google)

Basecut Cloud is an optional analysis engine included with paid Basecut plans. It sends content to Basecut's server function, which forwards it to Google's Gemini API using Basecut's own paid-tier Google account. Your Google key is not involved, and you do not need one. You can use Argus or your own key instead at any time.

When Basecut Cloud analyzes a clip, it splits the clip into sections (up to about 100 seconds each) and sends, per section:

• The section as a short, silent video taken from Basecut's reduced-size preview copy, not your original file, at most 720p (Basecut sends a smaller 480p copy instead when the preview is too large to send). If a section can't be sent as video, Basecut falls back to a small set of still frames (JPEG) from that preview instead, with the timestamp of each frame.
• The section's audio track (mono, up to about 100 seconds).
• Technical metadata: a guess at the camera model, the source color profile, video codec, source and section resolution, source and section duration, audio codec, and where the section falls within the longer clip.
• A motion level (none, low, medium, or high) that Basecut measures on your own computer before sending anything.
• A speech transcript excerpt, if you turned transcription on. The app can send up to 4,000 characters of a section's transcript to Basecut's server, and the server passes only the first 800 characters on to Google.

Whole-file selection. Basecut Cloud also has an optional ranking step that Basecut can switch on or off on its server. When it is on, after a file's sections have been analyzed, the app sends one text-only request per source file. For every section it contains the section's start and end time, its transcript (up to 1,500 characters per section), and a short note of what the earlier analysis saw and heard (up to 400 characters). The whole request is capped at 240,000 characters, so a very long file has its transcripts shortened to fit. This request carries no video, frames, or audio and no file names, but it can contain more of what was said in your footage than the analysis step does.

Basecut's analysis instructions are added on Basecut's server, not sent by the app: the app only sends the fields above, not free-form prompts. Your original video files are never uploaded; only the short, reduced-size sections above are sent, one at a time. For AI Looks, the app sends the text description you type, the previous look's numeric grade values if you ask to change a look, and short reference-look descriptions.

Auto-edit and the AI co-editor. These are optional Basecut Cloud features that help you build an edit. They never send your original files, video, audio, or file paths. What they send:

• Auto-edit requests. When you type a request, the app sends your message (up to 600 characters), a short summary of the current edit if there is one (its length, number of shots, where the music sits, and the scene names in it with how long each runs), the dates you filmed on, and the names of the songs it can use. Song names come from the file names of the songs in your music folder.
• Choosing what to keep. To judge which stretches of talking are worth keeping, the app sends up to six short transcript excerpts at a time, each with the lines just before and after it and a short description, plus up to three small still pictures from that moment (at most six pictures per request, each no larger than 768 pixels on its longest side).
• Picking shots. To score candidate shots, the app sends up to twelve small pictures per request, each a sheet of still frames from one shot no larger than 1,024 pixels on its longest side, with a short label. The label is Basecut's description of the shot or, for media you add to an auto-edit yourself, the label you gave it or, if you gave none, its file name.
• The AI co-editor. The app sends a text description of your timeline: the sequence's name, length and frame size, its tracks, and for each clip (up to 150) its position, length and name, Basecut's notes and rating for that shot, and short transcript excerpts, plus transitions, up to 30 marker names, and the instruction you typed, if any (up to 600 characters). A clip's name is often its file name. The co-editor sends no pictures, video, or audio.

The pictures and excerpts come from your footage, so they can show the people in it and include what they said. These requests take the same route as clip analysis, described next: the server does not store or log your messages, pictures, excerpts, timeline details, or results, and it keeps the same usage record for each request. Auto-edit also has its own hourly and monthly request limits and a monthly spending limit inside your plan's allowance, so it cannot use up the allowance clip analysis needs.

The route: the app sends this to Basecut's server function (hosted on Supabase), which forwards it to Google's Gemini API using Basecut's own paid-tier API account, model Gemini 3.8 Flash, and returns the result to the app. Sections are sent inline with the request, not uploaded to Google's separate file storage service. The server function does not store the frames, video, audio, prompts, or results.

What Basecut does store, per request: the account id, time, model name, token counts, computed cost, request type (for example clip analysis, selection, look, co-editor, or auto-edit), an optional run id (for co-editor requests, the name of the import, which can be your footage folder's name, cut to 64 characters), status, the device id of the computer that made the request, and a per-request random retry key used to stop duplicate charges (not linked to your content). This is used to enforce the monthly allowance, prevent abuse, and show the “clips left” meter. Our retention policy for these usage records is 24 months, and they are deleted together with your account if you delete your Basecut account.

Google is our sub-processor for Basecut Cloud. On the paid tier, Google does not use this content to train or improve its products. Google may retain it for a limited period as described in Google's Gemini API terms, for abuse monitoring and legal compliance.

The monthly allowance is a fair-use limit per paid account per calendar month (UTC). When you reach it, Basecut Cloud stops until the next month: if Argus is installed, the import carries on with Argus on your computer; otherwise analysis pauses, with nothing deleted or marked failed, until the allowance resets or you add your own key.

Engine 3: Your own Gemini key (straight to Google)

If you save your own Google Gemini API key in Settings and switch it on, the app talks to Google directly from your computer using that key. The request does not pass through Basecut's servers, and Basecut does not receive, log, or keep a copy of the content. Basecut Cloud limits and usage records do not apply. Per section, the app sends Google:

• A silent video taken from Basecut's reduced-size preview copy, not your original file. Small previews go inside the request. Larger ones are uploaded to Google's file storage service for Gemini first.
• The section's audio, as a mono 16 kHz recording.
• A text prompt with technical details about the clip (camera guess, color profile, codec, resolution, duration), a motion level, the names of objects Basecut detected in sample frames, and, if transcription is on, a speech transcript excerpt of up to 800 characters.

For AI Looks, the app sends the description you type, any reference-look descriptions, and the previous look's grade values if you ask to change a look. When you save a key, the app checks it by asking Google to list the available models, which analyzes nothing.

Your key is used only with the Gemini model you picked in Settings, and Basecut never switches a run on your key to a different model. If that model is busy or has used up its daily allowance, the app waits and retries the same model, and if it still cannot answer, the import stops with a message so you can try again or finish it later.

Google's terms for your key and plan decide how Google handles this content, including how long it keeps it and whether it may use it to improve its products. Please read them for the plan your key is on.

Traffic every engine shares

These requests happen whichever engine you use, including Argus. None of them contain your footage, frames, audio, or transcripts. Like any internet request, each one reveals your IP address to the server that receives it.

Account and sign-in

When you sign into Basecut, it checks whether your account has active access. Authentication and your account record (email, plan, access status) are handled by Supabase, our accounts provider. Your purchase and subscription status is kept current by Dodo Payments, our payment processor, through a secure webhook. If you arrived at our site from an ad or a link, we may also store the campaign parameters from that first visit (for example utm_source) on your account record, so we know which channels actually work.

Device seat

Your plan includes a limited number of computers. When you are signed in, the app registers this computer and checks in with Basecut's server, sending a random device id generated on install, the computer's name (its hostname, for example “Sam's MacBook”), and whether it is a Mac or a Windows PC. This is how the device list in your account is shown, and it is how a seat limit is enforced. It does not depend on the diagnostics setting.

Updates

The app fetches a static /version.json file from this site (with a backup copy on a second Vercel address) to tell you when an update is available, and downloads the installer for updates. The request hits our hosting provider (Vercel) and is subject to their default access logs (IP, timestamp, user agent). We do not add tracking, fingerprinting, or analytics to this endpoint, and we do not associate the request with your license key.

Model and software downloads

Some features need files the app does not ship with: the Argus model (from Hugging Face, through the permission step described under Argus), the local engine software (from GitHub), and the speech-transcription model (from Hugging Face). These are downloads to your computer. They do not upload your content.

App diagnostics

The desktop app sends diagnostics to Basecut's server (hosted on Supabase). This is on by default so we can catch problems and see how the app is used, and you can turn it off at any time in Settings. What is sent while it is on:

• Identifiers. A random install id created on your computer, a random id for each time the app is opened, the app version, whether it is a Mac or a Windows PC, and the operating system version. The record itself never contains your name, email, account id, license key, or sign-in token: they are removed from every diagnostic record before it is sent. Whether an event is also linked to your account depends on whether you are signed in, as described next.
• Usage events. Counts and outcomes such as: the app was opened, an analysis started, finished, failed or was cancelled (with the engine used, the number of clips, how long it took, and an error type), an update was downloaded or installed, Argus was downloaded, an export finished, a checkout was opened.
• Run summaries. After an analysis: clip and section counts, total duration, timings, camera and codec names, the engine and model used, and how many clips got each verdict.
• Feedback you give. A thumbs up or down, or a rating you change, including its short reason or note if you type one, which clip section it was (its start and end time), and a one-way code made from the file name, never the name itself. AI Look descriptions you type are also included. Email addresses are removed from this text, and phone numbers are removed from rating notes.
• Hardware summary at launch. So we can see which computers cannot run Argus, the app sends a coarse summary: processor or chip model, processor architecture, total memory rounded to whole gigabytes, the graphics card name (on Windows PCs) and, for NVIDIA cards, its dedicated video memory rounded to whole gigabytes, free disk space on the drive that holds your library (rounded to whole gigabytes), and whether your computer meets the requirements to run Argus and, if not, whether that is the hardware or the memory. It contains no serial numbers, hardware ids, network addresses, computer name, user name, or file paths.

Diagnostics never contain your footage, frames, audio, transcripts, or file names. The only free text in them is what you type yourself into a feedback note, a rating note, or an AI Look description.

Signed in or signed out

When you are signed in to your Basecut account, the app sends each diagnostic event together with your account sign-in, to Basecut's own server only. The server uses it to mark the event as coming from a signed-in account, so we can tell real usage by account holders apart from unverified events (anyone can send an event with information that is public). That mark is what is stored with the event. When you are signed out, events are sent without any sign-in and stay anonymous. Turning diagnostics off stops them either way: nothing is sent, so nothing is linked to your account.

Crash reports

If the app or an analysis crashes, Basecut writes a crash report on your computer. It is not sent automatically. When you next open the app, a dialog offers to copy the report, show you the file, or send it to Basecut. It is only sent if you click Send, and that works even when diagnostics are turned off, because the click is your consent for that one report. The report has the app version, your operating system and processor type, the crash code, the name of the analysis run, and the last lines of the app's log. Before it is written, the app removes email addresses, keys and tokens, and your user name, shortens file paths to just the file name, and replaces your clip file names with a one-way code. This scrubbing is automatic and best-effort, so you can open the report and read it before deciding whether to send it.

What turning diagnostics off does and does not do

• It stops: usage events, run summaries, feedback events, the hardware summary (it is not even collected), and the Argus download events, whether you are signed in or not.
• It does not stop: sign-in and the access check, the device seat check, update checks and downloads, Argus, speech-model and software downloads, and any Basecut Cloud or own-key analysis you run. Those are needed for the app to work and are described above. Basecut Cloud's usage records are part of the Cloud service, not diagnostics.
• It does not stop a crash report you choose to send.
• The app still keeps a diagnostic log file on your computer while diagnostics are off. It is never sent while the setting is off.

This website

The basecut.app site is hosted on Vercel, which captures standard server access logs (IP, request path, user agent, timestamp) for every request. Beyond that, the website (not the app) uses a small set of analytics and advertising tools so we can see whether the site works and whether our ads are worth running:

• Vercel Analytics and Speed Insights: cookieless, aggregate page-view and performance metrics.
• Google Analytics 4: page views and site events (for example, clicking a download button). Uses cookies.
• Meta Pixel: lets us measure ads we run on Facebook/Instagram and show ads to people who visited this site. Uses cookies.
• Reddit Pixel and Microsoft Clarity: same purpose for Reddit ads and anonymized session heatmaps; only active if we run those campaigns.
• OpenAI ads measurement pixel: tells us whether an ad we run in ChatGPT led to a sign-up, checkout or purchase. Customer details it reads are hashed in your browser before they are sent. It does not load for visitors in the EU, EEA, UK or Switzerland.

These run only on this website, never inside the Basecut app, and none of them ever touch your footage. Standard browser tracking blockers will block them and the site works fine regardless. We do not sell personal information.

Payments

All purchases run through Dodo Payments, our merchant of record. Dodo collects the information needed to process payment and grant access: name, email, billing details, and payment method. We receive your name, email, and order info to provision your account. Card numbers and full payment details are handled by Dodo Payments and their payment processors and never reach us.

Email

If you contact us (at support@basecut.app or by replying to your receipt), we retain that correspondence so we can support you. Our support inbox is hosted by Google Workspace. We do not add you to any email list for this. We do not sell or share email addresses.

If you leave your email on a guide page for that camera's printable wedding-day cheat sheet, we store the address, show the download on the page, and email you a copy of the sheet. After that we send at most two short follow-up emails: one color tip for your camera and one about how Basecut works. Every email has a one-click unsubscribe link, and replying “unsubscribe” works too. Emails are sent from our Gmail account (lettuceatkansas@gmail.com), so Google processes them as our email provider. If you leave your email for something not released yet (the culling demo video or a launch notice), we store it to send you that one thing. Either way: never sold or shared.

If you make a Basecut account, we may send you up to five short emails about getting started: a welcome, help if setup seems stuck, a note when your free exports run low, a reminder before your trial ends, and a check-in if checkout did not go through. They stop once you buy a plan, each one has a one-click unsubscribe link, and they come from the same Gmail account. To decide whether we may send them where you live, we keep the country your sign-in or checkout came from (a two-letter code, never your IP address). We do not send these emails to accounts in the EU, EEA, UK or Switzerland, or when the country is unknown.

Your rights

You can request access to, correction of, or deletion of any personal information we hold about you (essentially: your email and order record) by emailing support@basecut.app or replying to your receipt. Footage and analysis results live on your machine, so deleting them is a matter of deleting the Basecut folders on your disk; we cannot do it for you. If you used Basecut Cloud or your own Gemini key, Google handles the content you sent as described above.

Changes

If we change this policy in a material way, we'll update the date at the top and post a note in the next release. Continued use of Basecut after a change means you accept the updated policy.

Contact

Privacy questions: email support@basecut.app and a human answers.