# Eddy docs (all pages) Index: https://eddystudio.ai/llms.txt # Eddy docs > How Eddy Studio, the agentic video editor for Mac, works, for creators and for the AI agents they use. Source: https://eddystudio.ai/docs/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build Eddy Studio is a Mac app that edits your talking-head and screen recordings for you. You drop in raw footage, and Eddy cleans the sound, cuts the retakes, adds captions and motion graphics, and renders a long video plus up to 3 Shorts. You ask for changes in chat, and most of what Eddy changes can be undone ([what can't](https://eddystudio.ai/docs/working-with-eddy/#undo-and-apply)). These docs describe what the app does today. Where something isn't confirmed yet, the page says **Unknown**, **Not yet** or **Not final** instead of guessing. The tables of agent tools, AI Tools, caption styles, edit stages and Mac requirements are generated from the app's source code, and the docs won't build if a page disagrees with the app. > **Private beta:** Eddy is in a private beta. The first 10 creators are hand-picked and free. There's a 30-second fit check at the bottom of this page, right before the link to apply. ## The proof LAB-0024, from the founder's own YouTube show, edited with the engine that ships in the Eddy beta: the long video and one of its Shorts. Nothing loads from YouTube until you click play. [Video on YouTube: LAB-0024, the long video (14:07), on YouTube](https://www.youtube.com/watch?v=eIGJmUUqUr8) [Video on YouTube: A Short cut from LAB-0024 (1:19)](https://www.youtube.com/shorts/sIuka-k1Hf0) ## Start here | If you want to | Read | | --- | --- | | Check your Mac and install Eddy | [Get started](https://eddystudio.ai/docs/get-started/) | | See each step from raw footage to finished files | [How an edit works](https://eddystudio.ai/docs/how-it-works/) | | See how far one real recording got, and what's still unmeasured | [A real recording](https://eddystudio.ai/docs/real-edit/) | | Change an edit, undo it, or use the one-click AI Tools | [Working with Eddy](https://eddystudio.ai/docs/working-with-eddy/) | | Set your colours, lettering, logo and captions | [Your look and style](https://eddystudio.ai/docs/your-look/) | | Let ChatGPT, Claude or another agent drive Eddy | [For agents](https://eddystudio.ai/docs/for-agents/) | | Paste one prompt and have your agent run an edit | [Agent quickstart](https://eddystudio.ai/docs/agent-quickstart/) | | Know what leaves your Mac | [Privacy](https://eddystudio.ai/docs/privacy/) | | See the limits and open questions | [Limits and unknowns](https://eddystudio.ai/docs/limits/) | | Look up a word these docs use | [Glossary](https://eddystudio.ai/docs/glossary/) | ## For AI agents If you are an agent, start at [/llms.txt](https://eddystudio.ai/llms.txt). It lists every page, and each page has a plain Markdown copy at the same address ending in `.md`. [/llms-full.txt](https://eddystudio.ai/llms-full.txt) has all of them in one file. ## What Eddy is not - It is not a cloud editor. Your footage stays on your Mac. - It never publishes for you. Nothing goes to YouTube unless you click Upload yourself, and whether that upload works in the beta is [still unknown](https://eddystudio.ai/docs/limits/). - It is not for Intel Macs or Windows. ## 30-second fit check Three questions before you apply. Nothing is saved or sent. 1. Is your Mac Apple silicon (M1 or later), on macOS 13 or later, with 16 GB of memory? Yes: Good. Eddy runs on that Mac and your footage stays there. No: Eddy won't run on an Intel Mac or on macOS older than 13. With less memory it warns you that edits may be slow or stop. [See the full list](https://eddystudio.ai/docs/get-started/#what-you-need). 2. Do you pay for ChatGPT or Claude? Yes: Good. Eddy edits through that plan. You sign in once, in your browser; Eddy never uses an API key. No: Eddy needs one of them to edit. Either one is enough. 3. Are you OK with an edit taking hours, not minutes? Yes: Good. Eddy keeps your Mac awake while it works. How long an edit takes is not measured yet, so plan for hours. No: Eddy isn't for fast turnarounds yet. [See how an edit works](https://eddystudio.ai/docs/how-it-works/#how-long-it-takes). [Apply for the beta](https://eddystudio.ai/#apply) --- # Get started > What your Mac needs, which accounts you need, and what the first launch looks like. Source: https://eddystudio.ai/docs/get-started/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build ## What you need | Need | Detail | | --- | --- | | Mac | Apple silicon (M1 or later). Intel Macs are not supported. | | macOS | macOS 13 or later. Eddy tells you if your Mac is older. | | Memory | 16 GB. With less, Eddy warns that edits may be slow or stop. | | Fast transcription model | Optional. About 5.6 GB, downloaded once when you click Download. It needs macOS 14 or later. Without it, Eddy transcribes with the model that comes with the app, which is slower on a long recording. | | Vision model | Optional. About 3.9 GB, downloaded only when you click Download on its card in the Doctor sheet. The vision model needs macOS 14 or later. Without it, Eddy labels scenes from on-screen text and screen movement. | | AI account | A ChatGPT or Claude subscription. Either one is enough. | | Internet | Needed to talk to ChatGPT or Claude and for the one-time model downloads. | | Free disk space | Depends on your footage. Before an edit starts, Eddy works out roughly how much it needs and tells you. Long 4K recordings can need tens of GB. | _Generated from the app's source code each time these docs are built. The rows after Vision model are written by hand._ > **Unknown:** Which macOS versions have been tested end to end is still being confirmed. macOS 13 is the minimum the app accepts. Eddy doesn't have its own account. You sign in with ChatGPT or Claude, and the sign-in opens in your browser. You never need a terminal. ## Install Eddy comes as a signed and notarized disk image for Apple silicon. Open it and drag Eddy to Applications. After that, Eddy updates itself (see [Updates](#updates)). > **Not yet:** There is no public download. During the private beta, Eddy is only for the creators the team picks. ## The first launch There is no welcome tour. The first screen is the drop zone where your footage goes. Beside it is a short **Get ready** list. None of its rows blocks you; you can drop footage straight away. 1. **Connect ChatGPT or Claude.** The editor runs on whichever one you sign in with. 2. **Your look.** Your colours, lettering and logo for motion graphics. Eddy can read them from your channel. See [Your look and style](https://eddystudio.ai/docs/your-look/). 3. **Fast transcription.** Optional: a one-time download (about 5.6 GB) of a faster transcription model that runs on your Mac. The fast transcription model needs macOS 14 or later, so this row only shows on Macs that can run it. Without it, Eddy transcribes with the model that comes with the app, which is slower on a long recording. 4. **Install.** Only appears if one of the app's own health checks fails, and tells you what to fix. Scene labels need no extra download: Eddy labels scenes from on-screen text and screen movement. An optional vision model (about 3.9 GB) also lets Eddy look at each scene, still on your Mac. It is only downloaded when you click **Download** on its card in the Doctor sheet, and the vision model needs macOS 14 or later. The **Doctor** sheet shows every health check in one place. It is also where you connect agents, turn on remote access and send a support report. ## Recording inside Eddy Eddy can record for you: your screen (the whole display or one window), your camera, your microphone and your Mac's sound. It also records where your cursor goes and where you click, so it can zoom to the right place later. macOS asks you for Screen Recording, Camera and Microphone permission the first time. ## Updates Eddy checks for updates shortly after launch and then every few hours. It downloads new versions in the background and installs them when you quit, or when you click **Restart now**. It never installs while you are recording, and it never goes back to an older version. --- # How an edit works > Every step an edit goes through, from dropping in footage to the finished files. Source: https://eddystudio.ai/docs/how-it-works/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build An edit in Eddy is a single job that runs through fixed stages. The progress bar shows each stage as it really happens. It doesn't show made-up percentages or a countdown. ## Before you start: what to make When you drop footage in, Eddy asks what it should make: | Choice | What you get | | --- | --- | | Long + 1 Short | One long video and one Short | | Long + 3 Shorts | One long video and up to 3 Shorts | | Short only | One Short | | Customize | Your own mix, up to 3 Shorts | Shorts are vertical 9:16 clips cut from your footage. The long video is 16:9. ## From raw footage to finished files Six steps, in order. Each one says which stage of the app's progress bar it happens in, and why it matters to you. 1. **Import** (during the app's Import stage). You drop in your raw recording: a camera file, a screen recording, or a folder with both. Eddy never changes your original files, and it cleans the sound (Studio Sound) as the footage comes in. Why it matters: Every later step works from the cleaned-up version of your voice. [Media: Import: the app's own screen at this step, from one real project.](https://eddystudio.ai/docs/media/pipe-import.webp) 2. **Transcript** (during the app's Import stage). Speech models on your Mac write down every word you said, with the time each word starts and ends. Why it matters: Cuts, captions and Shorts are all made from these word timings, so edits land between words instead of in the middle of one. [Media: Transcript: the app's own screen at this step, from one real project.](https://eddystudio.ai/docs/media/pipe-transcript.webp) 3. **Cuts** (during the app's Rough cut stage). Eddy removes retakes and shortens long pauses. Every cut keeps the reason it was made. Why it matters: You can see why each piece went, and bring any of it back with Undo. [Media: Cuts: the app's own screen at this step, from one real project.](https://eddystudio.ai/docs/media/pipe-cuts.webp) 4. **Scenes** (starts during the app's Import stage, in the background). In the background, Eddy labels each scene from what is on screen: it reads the text on screen and measures how much the screen moves. With the optional vision model installed, it also looks at each scene. Why it matters: The labels decide the layout, such as when your screen fills the frame and when you do. [Media: Scenes: the app's own screen at this step, from one real project.](https://eddystudio.ai/docs/media/pipe-scenes.webp) 5. **Captions** (during the app's Edit stage). In the edit stage, the AI editor you signed in with adds captions, motion graphics in [your look](https://eddystudio.ai/docs/your-look/), and the Shorts. Why it matters: Captions follow the transcript word by word, so the highlighted word is the one you're saying. 6. **Render** (during the app's Render stage). Eddy renders the long video and each Short as 1080p MP4 files. Then the app's Check stage scores the edit (and can fix the biggest gaps and render once more), and its Deliver stage copies the files to your folder. Why it matters: Nothing is published. The files are on your Mac, and what happens next is up to you. [Media: Render: a frame of a Short in the app's preview, from one real project. No file was exported for these docs.](https://eddystudio.ai/docs/media/pipe-render-short.webp) ## The stages These are the stages the progress bar shows, in order. | Stage | What Eddy does | | --- | --- | | **Import** | Cleans the sound (Studio Sound), writes the transcript with word timings, makes an editorial pass over what you said, and starts labelling the scenes in the background. | | **Rough cut** | Removes retakes and shortens long pauses. | | **Edit** | The AI editor builds the story, the hook, motion graphics, captions and the Shorts. | | **Render** | Renders the long video and each Short. | | **Check** | Checks the long video against Eddy's quality bar. If it falls short, Eddy makes one round of fixes and renders again. A low score never blocks delivery. Skipped when you only asked for a Short. | | **Deliver** | Copies the finished files to your folder and checks that each copy is complete. | _Stage names are generated from the app's progress bar each time these docs are built._ When the last stage is done, the edit shows as ready. Nothing is published. Uploading is your own click in Eddy. ### Studio Sound Eddy's sound clean-up (called Studio Sound in the app) is always on. It is applied when the footage comes in, and there is no switch for it. ### Retakes and pauses Eddy finds the takes you restarted and keeps one take, and it shortens long gaps between words. Every cut carries the reason Eddy made it (a retake, a filler word, a stutter, a long pause and so on), so you can see why something was removed and put it back. ### Scenes Eddy labels the scenes from what is on screen. It reads the text on screen and measures screen activity on your Mac. If you install the optional vision model from the Doctor sheet (the vision model needs macOS 14 or later), it also looks at each scene, still on your Mac. You can let ChatGPT or Claude look at a few still frames for a project too, but only if you switch that on for that project. ### The edit itself The editing stage is done by the AI editor you signed in with. Its brief asks for: - an opening 30 seconds that works as a hook, - captions on the long video and on the Shorts, - motion graphics in [your look](https://eddystudio.ai/docs/your-look/), with regular movement on Shorts, - sound effects from Eddy's own library, - screenshots of web pages you mention, taken from public pages only. ## What you get This is the standard the beta is held to: LAB-0024 from the founder's own YouTube show, edited with the engine that ships in the Eddy beta. A long video and a Short from one recording. Click a video to play it; nothing loads from YouTube until you do. [Video on YouTube: LAB-0024, the long video (14:07), on YouTube](https://www.youtube.com/watch?v=eIGJmUUqUr8) [Video on YouTube: A Short cut from LAB-0024 (1:19)](https://www.youtube.com/shorts/sIuka-k1Hf0) Each output is an MP4 at 1080p in high quality. The files are named after your footage, ending in `-long.mp4` and `-short-1.mp4`, `-short-2.mp4` and so on. By default they go to a folder for that video inside `Movies/Eddy Studio` in your home folder. If your footage is too short to carry as many Shorts as you asked for, Eddy makes fewer and says so when the edit is ready. If a Short it planned can't be made, Eddy tries to fix it. If it still can't, the edit stops and tells you why, rather than delivering an incomplete set. You can add a note and try again. From the **Export** sheet you can also export by hand: MP4, WAV or MP3 audio, or SRT and VTT caption files, at 1080p or 720p. ## How long it takes A full edit takes hours, not minutes. Exactly how long is **Not yet measured**: no beta test run has finished a real recording yet. [A real recording](https://eddystudio.ai/docs/real-edit/) shows the count. > **Not yet measured:** No beta test run has yet taken a real recording all the way to a finished long video and Short, so there is no measured time to publish. [A real recording](https://eddystudio.ai/docs/real-edit/) shows how far a real run got. Eddy keeps your Mac awake while it works, and asks you to plug in if you are on battery. Long recordings can sit in the transcript step for many minutes without moving the bar; that is normal. If your Mac sleeps or you quit, the edit can resume where it stopped. --- # A real recording > One real recording through Eddy: how far it got, what the app changed in it, and what hasn't been finished or measured yet. Source: https://eddystudio.ai/docs/real-edit/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build This page follows one real recording through Eddy. It shows only what the app actually made from it. Nothing here is a mock-up. > **Not yet:** A finished edit to show. The beta run on this recording stopped during the edit stage, when the app restarted for an update, before Eddy had planned the motion graphics. No beta test run has yet taken a real recording all the way to a finished long video and Short, so there is no finished file on this page, and no time. ## What a finished edit looks like Until a beta run finishes, the closest proof is LAB-0024, an episode of the same show, edited with the engine that ships in the Eddy beta. It is the quality bar the beta is measured against. It is not this recording. [Video on YouTube: LAB-0024, the long video (14:07), on YouTube](https://www.youtube.com/watch?v=eIGJmUUqUr8) [Video on YouTube: A Short cut from LAB-0024 (1:19)](https://www.youtube.com/shorts/sIuka-k1Hf0) ## The recording A real episode of the founder's own YouTube show: a 66-minute 1080p camera recording with sound, plus a 4K screen recording of the same session. It went into Eddy the normal way, by dropping in the folder. [Media: Before: a frame of the raw recording, untouched.](https://eddystudio.ai/docs/media/real-before.webp) ## What Eddy changed in it Each AI Tool was run on this recording in the app. These are the app's own words for what each one did: - **Remove retakes**: "Removed 28 retakes · 3:26 shorter · 10 to review" - **Shorten word gaps** (default settings): "Shortened 184 gaps · 37:35 shorter" - **Edit for clarity**: "Ignored 14 duplicated sentences · 0:12 shorter" - **Remove filler words**: "Nothing detected in this composition." - **Style all scenes** (no button; the chat editor and agents run it): "Styled 260 layers", though the preview looked the same before and after. [Working with Eddy](https://eddystudio.ai/docs/working-with-eddy/#ai-tools) has a before and after clip, cut from this recording, for each tool that made a change you can see. They are **Not published yet**: they go up once the founder has okayed them. ## How far it got These are frames from Eddy's preview of this project. The [pipeline on How an edit works](https://eddystudio.ai/docs/how-it-works/#from-raw-footage-to-finished-files) uses the same frames, step by step. Like the clips, they are **Not published yet**. [Media: A Short (9:16) in Eddy's preview, before the edit stage finished. No file was exported.](https://eddystudio.ai/docs/media/real-short.webp) ## How long it took > **Not yet measured:** No beta test run has yet taken a real recording all the way to a finished long video and Short. Until one does, there is no honest time to show, so this page doesn't show one. Eddy's beta test is five runs in a row on the same build, each from a real recording to the finished files, each timed by the clock. Passing runs so far: 0 of 5. When a run passes, its measured time and the finished files go here. --- # Working with Eddy > Changing an edit by chat, undoing changes, the one-click AI Tools, and sending the team feedback. Source: https://eddystudio.ai/docs/working-with-eddy/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build ## Ask in chat The chat panel is where you talk to the editor. Say what you want in plain words, like "show my screen while I list the models" or "cut the part where I restart". Eddy shows you what it changed. Most changes land straight away; a few [wait for you to press Apply](#undo-and-apply). - A chat that isn't tied to part of the video is the **Editor**. It can work on the whole video. - A chat you open on a scene, or on words you selected, is a **Sub-editor**. It can change only the scenes it was opened on and the layers in them, and anything else it tries is refused. Motion graphics are the one exception: they're tied to your words, not to a scene, so a Sub-editor can add them anywhere. - You can have several chats open in tabs. - The box where you type has a model picker and an effort setting. [Media: Before: Eddy's first edit zoomed in on the screen, but the camera was covering it.](https://eddystudio.ai/assets/media/m4-before.mp4) [Media: After: the creator asked in chat to show the screen while they listed the models, and Eddy moved the camera aside.](https://eddystudio.ai/assets/media/m4-after.mp4) ## Undo and Apply Most changes apply straight away, whether you ask in chat or click an AI Tool, and you can undo them: - A change that applied straight away shows a result line with its own **Undo**, which takes back exactly that change. - The header has Undo and Redo, and Cmd+Z works too. - Every cut keeps the reason it was made, such as a retake, a filler word, a stutter or a long pause. Undoing a cut puts the words back. - If a later change touched the same thing, undo the later one first. One kind of change isn't in Undo yet: when the editor rewrites its plan for the edit (such as the sound plan or which Shorts to make), the new plan replaces the old one. Some chat changes wait on a card until you press **Apply** (or **Reject**): - a change that deletes a layer; - a change from the **Editor** that adds or removes scenes, or changes the timing, layout, transition or design of scenes, in more than one scene at once (changes to layers, cuts and motion graphics don't count); - a style measured from a reference video you pasted into chat; - a change the editor asks you to check first. A change you applied from a card has no Undo of its own. Use the header's Undo or Cmd+Z. Instant changes only apply inside your project on your Mac. Uploading to YouTube only happens when you click **Upload** yourself. Editing itself does send your transcript, other relevant text and a few selected frames to ChatGPT or Claude; [Privacy](https://eddystudio.ai/docs/privacy/) lists exactly what goes where. ## AI Tools The AI Tools bar has one-click actions for common jobs. **Style all scenes** has no button yet: ask for it in chat, or have an agent on the same Mac run it. | Tool | Where it is | What it does | Can cut parts of the video | | --- | --- | --- | --- | | **Edit for clarity** | Button in the AI Tools bar | Finds whole sentences you said more than once and keeps only the last one. It ignores capitals, punctuation and symbols when it compares them, so two sentences that differ only in those (like $1.50 and $150) count as the same | Yes | | **Remove filler words** | Button in the AI Tools bar | Takes out filler words | Yes | | **Remove retakes** | Button in the AI Tools bar | Cuts the false starts, restarts and repeated phrases Eddy marked while reading your recording | Yes | | **Shorten word gaps** | Button in the AI Tools bar | Shortens long pauses between words. It never cuts a word | Yes | | **Style all scenes** | No button: ask in chat, or an agent on the same Mac runs it | Applies one look (corner rounding or a colour preset) to every layer of one kind, such as every camera shot | No | | **Studio Sound** | Always on in the AI Tools bar, no button | Applied automatically when the recording is imported (the app's own words) | No | _Tool names, where each one is, and the last column are generated from the app's source code each time these docs are built. "Yes" means the tool can cut parts of the video, which makes it shorter. Every cut can be undone._ **Shorten word gaps** shortens any pause longer than 0.3 seconds down to 0.1 seconds by default, and you can change both numbers before you run it. Pauses Eddy already marked as long while reading your recording are the exception: those are shortened the way Eddy's own edit shortens them, whatever the two numbers say. Like chat, each one applies at once and can be undone. [Media: The AI Tools panel in the app, right after **Remove retakes** ran on a real recording. The result line and its **Undo** are the app's own.](https://eddystudio.ai/docs/media/ai-tools-panel.webp) ### Each tool on a real recording Each tool below was run in the app on the same real recording as [A real recording](https://eddystudio.ai/docs/real-edit/). Where there is a clip, the "after" is the same stretch of the recording with exactly the cuts the tool made, and nothing else changed. #### Remove retakes On the whole recording, the app said: "Removed 28 retakes · 3:26 shorter · 10 to review". [Media: Before: 18.0 seconds of the raw recording](https://eddystudio.ai/docs/media/ai-remove_retakes-before.mp4) [Media: After **Remove retakes**: the same stretch, now 12.7 seconds](https://eddystudio.ai/docs/media/ai-remove_retakes-after.mp4) #### Remove filler words > **Unknown:** What **Remove filler words** does to a real recording. On this one the app said "Nothing detected in this composition.", so there is no before and after to show yet. #### Shorten word gaps On the whole recording, with the default settings, the app said: "Shortened 184 gaps · 37:35 shorter". [Media: Before: 17.5 seconds of the raw recording](https://eddystudio.ai/docs/media/ai-shorten_gaps-before.mp4) [Media: After **Shorten word gaps**: the same stretch, now 9.6 seconds](https://eddystudio.ai/docs/media/ai-shorten_gaps-after.mp4) #### Edit for clarity On the whole recording, the app said: "Ignored 14 duplicated sentences · 0:12 shorter". [Media: Before: 10.0 seconds of the raw recording](https://eddystudio.ai/docs/media/ai-edit_for_clarity-before.mp4) [Media: After **Edit for clarity**: the same stretch, now 9.5 seconds](https://eddystudio.ai/docs/media/ai-edit_for_clarity-after.mp4) #### Style all scenes > **Unknown:** What **Style all scenes** looks like on a real recording. On this one the app said "Styled 260 layers", but the preview looked exactly the same before and after, so there is nothing honest to show yet. ## Send feedback (the improvements box) The **Feedback** button at the top of the chat panel is how you tell the team what to improve. Write a note in your own words (up to 4,000 characters) and press Send. - Your note goes with your app version and basic Mac details (macOS version, chip and memory). The form shows these under the box before you send. - You can tick **Add my anonymised session log** to include recent edit stages, crash notes and log lines. If you tick it, you see the whole thing before it goes, and Send sends exactly what you saw. - No footage, files, transcripts or chats are sent. - Before anything is sent, Eddy removes your home folder path, user name, email addresses and keys from it. - If Eddy can't reach the team, it keeps the note on your Mac and tells you where. Notes are stored privately for the Eddy team. The team reads them, and also uses automated AI runs to sort them into its own private task list. ## Support reports If something goes wrong, the Doctor sheet has **Send a report to Eddy**. It sends your app and Mac details, the health checks, the stages and errors of your recent edits, and recent log lines, all cleaned the same way as feedback. You see the full report before you send it. --- # Your look and style > Your colours, lettering and logo, caption styles, copying a format you like, and the house style. Source: https://eddystudio.ai/docs/your-look/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build ## Your look **Your look** is your colours, lettering and logo. Eddy uses it on every motion graphic it makes. You set it from the **Get ready** list, and Eddy can fill it in from your channel. You can change it at any time, or ask an agent to change it for you. ## Captions Eddy's edit puts captions on the long video and on the Shorts. The Captions panel has 4 styles. **House** is the default: bold white words, up to 5 at a time, with the word being spoken in red. | Style | Words on screen at once | Lettering | Text / spoken-word colour | | --- | --- | --- | --- | | **House** (default) | 5 | Arial, bold | #FFFFFF / #E5484D | | **Clean** (not final) | 4 | Helvetica Neue | #FFFFFF / #4AA3FF | | **Classic** (not final) | 4 | Arial | #FFFFFF / #FFE24D | | **Bold** (not final) | 3 | Futura, bold, capitals | #FFFFFF / #E5484D | _Generated from the app's source code each time these docs are built._ > **Not final:** Clean, Classic and Bold are still provisional in the app. How they look may change during the beta. In the panel you can also change the colours and the position, and choose which part of the video a change applies to. You can export the captions on their own as SRT or VTT files from the Export sheet. > **Unknown:** A single on/off switch for captions isn't confirmed in the current app. If you don't want captions, ask in chat and check the result. ## Copy a format you like You can give Eddy a reference, such as a link to a YouTube video or Short, a reel, a TikTok, or a video file on your Mac, and ask it to copy the format. Eddy downloads the reference to your Mac and measures how it is edited: the pacing of the cuts, the pauses, the layouts, the zooms and the captions. It saves that under a name, so you can ask for it again on a later video. A format is timing and structure only. Eddy does not keep the reference's colours, fonts, logo, words, music or footage. Your graphics stay in your own look. Formats are saved on your Mac only. ## House style Unless you ask for one of your saved formats, an edit uses Eddy's built-in house style: its scene layouts, motion and timing. Ask for a format and Eddy matches that format's timing and structure instead (the pacing of the cuts, the zooms and the layouts), still in your own look. The house style is about how the whole edit looks and moves. It is not the **House** caption style, which is only how captions look. > **Not final:** The built-in house style is still being tuned and hasn't been signed off. Expect how edits look and move to change during the beta. > **Not yet:** A house style learned from your own videos. --- # For agents > How ChatGPT, Claude, Codex and other AI agents connect to Eddy and drive an edit, and what they can't do. Source: https://eddystudio.ai/docs/for-agents/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build Eddy is built to be driven by AI agents as well as by hand. An agent connects over MCP (the Model Context Protocol) and can do the whole job: check that Eddy is ready, find footage, record, start an edit, check on it, look at frames, and ask for changes. ## If you are an agent reading this - Every page here has a plain Markdown copy: add `.md` to the page address, e.g. [/docs/for-agents.md](https://eddystudio.ai/docs/for-agents.md). - [/llms.txt](https://eddystudio.ai/llms.txt) lists every page. [/llms-full.txt](https://eddystudio.ai/llms-full.txt) is all of them in one file. - Eddy runs on the creator's Mac. You can only use it while Eddy Studio is open there and the creator has connected you. - Never tell a creator a video is published or uploaded. Eddy can't do that for you; see [What agents can't do](#what-agents-can-t-do). - Want one prompt to paste into ChatGPT or Claude? See the [Agent quickstart](https://eddystudio.ai/docs/agent-quickstart/). ## Connect an agent on the same Mac In Eddy, open the **Doctor** sheet and use the **Connect your agent** card. It adds Eddy to Claude Desktop, Claude Code or Codex in one click. It only changes Eddy's own entry, and for Claude Desktop and Codex it backs up the settings file first. The card also shows the setting to paste by hand if you prefer. A local agent gets the creator tools below and also the full set of detailed editor tools, one per action in the app, for precise timeline work. ## Connect ChatGPT or claude.ai (remote access) Remote access is off until the creator turns it on in the Doctor sheet (**Use Eddy from ChatGPT and claude.ai**). 1. With remote access on, the Mac opens one outgoing connection to Eddy's relay at `relay.eddystudio.ai`. The Mac needs no open port. 2. In ChatGPT (developer mode connectors) or claude.ai (custom connectors), add the address the card shows. 3. A sign-in page opens and asks for the 8-character code shown on the Mac. The code works once and expires after 10 minutes, and each sign-in gets 5 tries. This shows the person signing in can see Eddy on the Mac. For remote access, the relay keeps the connected apps, their grants, scrambled (hashed) tokens and your Mac's network name (which often includes your name), plus the sign-in code for its 10 minutes. It does not store footage or the content of calls. Access tokens last an hour and refresh tokens rotate; a refresh token used twice ends the grant. The card lists every connected app with a **Remove** button, which cuts that app off at once. A remote agent gets the creator tools only, never the detailed editor tools. ## The creator tools There are 15 creator tools. Every agent gets them, local or remote. The descriptions are the app's own words, the ones your agent reads. | Tool | Name in the app | What it does (the app's own words) | Read-only | | --- | --- | --- | --- | | `eddy_get_started` | Get started with Eddy | Check Eddy is ready to edit before anything else: Studio running, the creator's AI editor signed in, fast transcription, their brand look, and edits in progress, with the next step. | Yes | | `eddy_sign_in` | Sign in the AI editor | Start signing Eddy in to the creator's own ChatGPT or Claude account, which Eddy edits with. | No | | `eddy_find_footage` | Find footage | Find the creator's recent recordings when they say "edit yesterday's recording" or don't know the path: Eddy recordings, Movies, Desktop and Downloads, newest first, each with the path for eddy_edit_this and its shape. | Yes | | `eddy_record` | Record | Record the creator's screen and/or camera with Eddy's native recorder when they say "record this": start, pause, resume, stop, status, or list the inputs. | No | | `eddy_edit_this` | Edit this footage | Edit this recording into a finished YouTube long, 1 to 3 Shorts, or both, when the creator says "edit this". | No | | `eddy_get_edit` | Check an edit | Check how an edit is going when the creator asks: stage, progress, a one-line status, and once ready the long and Shorts with their file paths. | Yes | | `eddy_watch` | Watch the finished edit | Look at the finished edit before describing it: frames of the long or a Short as images, at given seconds or spread evenly across it (a contact sheet). | Yes | | `eddy_refine` | Refine the edit | Change a finished edit by prompt when the creator asks ("make the intro punchier", "Short 2 needs a stronger hook", "cut the tangent at 4:10"). | No | | `eddy_resume_edit` | Resume an edit | Carry on an edit that failed or was interrupted (Eddy Studio closed) from the stage it stopped in. | No | | `eddy_stop_edit` | Stop an edit | Stop a running edit when the creator asks: its import, editing and rendering. | No | | `eddy_draft_video_details` | Write the video details | Write the YouTube title and description for the long or a Short of a finished edit, so Eddy's Export sheet opens with them filled in. | No | | `eddy_change_look` | Change the brand look | Set or refine the creator's brand look for motion graphics when they describe their brand or ask for a change ("punchier", "a little less bounce", "fewer graphics", "use my channel's colours"): fill it from their YouTube channel, set the name, colours, light or dark, lettering and how the graphics move. | No | | `eddy_list_projects` | List projects | List the creator's Eddy projects, newest first: name, length and when each was last opened. | Yes | | `eddy_set_up_loop` | Set up the creator's loop | Read, set up or change how this creator makes a video, from the plan to the release: what to call them, their channel and niche, their shows (most have one; each has a name, its video names, the days and time they film it, and can be paused for a break), where their footage comes from, Shorts per video, whether Eddy grills them from the transcript before the edit, their own skill for planning, editing or the release (or Eddy's built-in step), and what goes out beside YouTube. | No | | `eddy_read_transcript` | Read the transcript | Read what the creator said in a recording, as lines with the m:ss each starts at, before an edit: to grill them on the cut quoting their own words, or to plan the Shorts. | No | _Generated from the app's source code each time these docs are built._ A typical run: `eddy_get_started`, then `eddy_find_footage` and confirm the file with the creator, then `eddy_edit_this`, `eddy_get_edit` every few minutes until it's ready, `eddy_watch`, and `eddy_refine` for changes. An edit takes hours, not minutes. No measured time is published yet (see [A real recording](https://eddystudio.ai/docs/real-edit/)), so don't promise the creator a time. ## Ready-made prompts Eddy also offers MCP prompts. In apps with a prompt menu (Claude Desktop, for example) the creator can pick one instead of typing. | Prompt | Shows up as | What it starts | | --- | --- | --- | | `edit-this` | Edit this video with Eddy | Edit a recording end to end into a YouTube long plus Shorts, or Shorts alone. | | `set-up-my-loop` | Set up my video loop with Eddy | First run: set up how you make each video (plan, film, edit, review, release), with your own skills where you have them. | | `run-my-loop` | Run my video loop with Eddy | Take one video through your loop: plan it, film it, edit it once the footage is in, review it, release it. | _Generated from the app's source code each time these docs are built._ ## What agents can't do These stay with the person at the Mac: - **Publish or upload.** A YouTube upload only starts from the creator's own click in the Export sheet. No agent, AI Tool or background job can start one. - **Turn remote access on or off**, or add and remove connected apps. - **Sign in to YouTube** for the creator. - **Record the creator's verdict** on an edit, or resolve review comments. ## The local HTTP API Behind the app is a local HTTP API on the Mac. It needs a secret that changes every launch, and it only answers on the Mac itself. It is internal and may change without notice. Use MCP instead. --- # Agent quickstart > Connect ChatGPT or Claude to Eddy, paste one prompt, and let it run an edit with you. Source: https://eddystudio.ai/docs/agent-quickstart/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build Three steps: connect your agent to Eddy, paste the prompt below, then answer its questions. Eddy Studio has to be open on your Mac the whole time. ## 1. Connect your agent **ChatGPT or claude.ai in the browser** (remote access): 1. In Eddy, open the **Doctor** sheet and turn on **Use Eddy from ChatGPT and claude.ai**. 2. In ChatGPT, add Eddy as a developer mode connector. In claude.ai, add it as a custom connector. Use the address the Doctor card shows. 3. A sign-in page asks for the 8-character code shown on your Mac. It works once and expires after 10 minutes. A remote agent gets the 15 creator tools only. [More about remote access](https://eddystudio.ai/docs/for-agents/#connect-chatgpt-or-claude-ai-remote-access). > **Unknown:** Which ChatGPT and Claude plans allow adding your own connectors. Check your plan's settings. **Claude Desktop, Claude Code or Codex on the same Mac:** open the **Doctor** sheet and click your app in the **Connect your agent** card. [More about local agents](https://eddystudio.ai/docs/for-agents/#connect-an-agent-on-the-same-mac). ## 2. Paste this prompt It works the same in ChatGPT and in Claude. Fill in the two bits in square brackets, or delete them. ```text You're driving Eddy Studio, the video editor on my Mac, through its eddy_ tools. 1. Call eddy_get_started first and fix whatever it says is missing. If my AI editor isn't signed in, call eddy_sign_in and tell me to approve the page that opens on my Mac. 2. Find my footage with eddy_find_footage, or use this path: [path to my recording]. Tell me which file you picked and wait for my OK. 3. Read everything I said with eddy_read_transcript (a long recording comes in parts). Ask me up to three questions about the cut, quoting my words with their times, and wait for my answers. 4. Start the edit with eddy_edit_this: a long video plus [1 to 3] Shorts, with my answers, in my own words, as its notes. 5. Check it with eddy_get_edit every few minutes, not more often, and tell me when the stage changes. Don't promise a finish time: an edit takes hours, and no edit time has been measured yet. 6. When it's ready, look at it with eddy_watch before you describe it. Ask what I'd change, and pass my words to eddy_refine exactly as I said them. 7. Never tell me a video is published or uploaded. Uploading is my own click in Eddy. ``` In an app with a prompt menu, such as Claude Desktop, you can pick **Edit this video with Eddy** instead. It asks for the path to your footage and how many Shorts you want. ## 3. While it runs - Leave Eddy Studio open and your Mac on. Eddy keeps the Mac awake during the edit. - If the agent says the edit stopped, ask it to call `eddy_resume_edit`. The edit carries on from the stage it stopped in. - When it's done, your files are in your folder (see [What you get](https://eddystudio.ai/docs/how-it-works/#what-you-get)). Nothing is uploaded. ## How we test these docs on an agent A fresh AI agent that had never seen Eddy was given only this website and the 10 questions below. Its answers were marked against an answer key written from the app's source code before the run. **Score: 10 out of 10** (3 November 2026, on the docs as they are now). 1. What Mac, macOS version and memory does Eddy need? 2. What does Eddy download on first use, and how big is each download? Does either one have its own macOS requirement? 3. Which AI account does Eddy need, and where does a creator apply for the beta? 4. How long does an edit take, and has that time been measured? 5. Which MCP tool should an agent call first, and which tool starts an edit? 6. Can an agent publish or upload the finished video to YouTube? 7. How many Shorts can one edit make, and what shape are they? 8. Which one-click AI Tool removes repeated takes, and can its changes be undone? 9. What is the default caption style, how many words does it show at once, and is Eddy's built-in house style final? 10. How does ChatGPT or claude.ai in the browser connect to Eddy, and which tools does a remote agent get? If your agent gets one of these wrong after reading the docs, tell us at [support@eddystudio.ai](mailto:support@eddystudio.ai): the page needs fixing, not your agent. --- # What leaves your Mac > What stays on your Mac, what goes to your AI provider, and what reaches the Eddy team. Source: https://eddystudio.ai/docs/privacy/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build This page explains the app's behaviour in plain words. The [privacy policy](https://eddystudio.ai/privacy/) is the formal version, and it wins if the two ever differ. ## Stays on your Mac - Your footage. Eddy never changes your original files; edits are made on top of them. - Your finished videos, until you choose to do something with them. - The transcript work, which runs on your Mac with the speech model that comes with the app, or the optional fast transcription model if you download it. - Scene labelling, which reads the screen on your Mac, and the optional vision model if you install it. - Your saved formats. ## Goes to your AI provider To edit, Eddy sends the transcript, other relevant text and a few selected frames to ChatGPT or Claude, through your own account. That provider's terms apply to what it receives. For scene labels, you can also let ChatGPT or Claude look at a few still frames. That is off unless you turn it on for a project. If you connect ChatGPT or Claude as an agent (see [For AI agents](https://eddystudio.ai/docs/for-agents/)), that agent also receives what its Eddy tools return, such as the transcript, frames of your edit and file names. That provider's terms apply to it too. ## Reaches the Eddy team **Anonymous usage, on by default.** The app sends the team anonymous usage events and a masked screen replay of the app window. The events say which buttons and screens you use, how long each step takes, which tools Eddy ran and whether they worked. In the replay your video, pictures, words and file names are blanked out. It never sends your footage, audio, transcript, chat or file names. It goes to PostHog's US cloud under a random id, not your name or email. Turn off **Send anonymous usage** in the Doctor sheet and it stops at once. Everything else, only when you choose to send it: - a note from the **Feedback** box ([what it includes](https://eddystudio.ai/docs/working-with-eddy/#send-feedback-the-improvements-box)), - a **Send a report to Eddy** support report from the Doctor sheet, - crash reports, which are off unless you turn them on in the Doctor sheet. Feedback notes and support reports are cleaned first: your home folder path, user name, email addresses and keys are removed. Crash reports are different: they are the app's crash dumps, sent as they are, and can include bits of the app's memory such as a file name. Notes and reports are stored privately for the team. Besides the team reading them, automated AI runs sort them into the team's private task list. ## Other connections - **Updates.** The app asks `updates.eddystudio.ai` whether a new version exists. - **Remote access.** Only if you turn it on. Your Mac connects out to Eddy's relay so ChatGPT or claude.ai can reach it. The relay passes calls through and stores no footage. See [For agents](https://eddystudio.ai/docs/for-agents/#connect-chatgpt-or-claude-ai-remote-access). - **Model downloads.** The optional fast transcription model and the optional vision model, if you install them, are downloaded once from Hugging Face. The vision model's software comes from PyPI. - **Web page screenshots.** When the edit uses a web page you mention, Eddy loads that public page to take its picture. - **Reference videos.** When you ask Eddy to copy a format, it downloads that reference video to your Mac. - **Your channel and YouTube.** Reading your look from your channel, and any YouTube upload, only happen when you ask. --- # Limits and unknowns > What Eddy doesn't do yet and what these docs couldn't confirm. Source: https://eddystudio.ai/docs/limits/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build Eddy is in a private beta. This page lists what it doesn't do and what these docs couldn't confirm, so you and your agent don't promise more than the app does. ## Doesn't do - **Intel Macs, Windows or the web.** Eddy is a Mac app for Apple silicon only. - **Work fully offline.** It needs your ChatGPT or Claude account to edit. - **Publish for you.** Uploading is only ever your own click. - **Fast turnarounds.** A full edit takes hours. - **More than 3 Shorts** from one edit. ## Not confirmed yet > **Unknown:** Which macOS versions are fully tested. The app accepts macOS 13 or later. > **Not yet measured:** How long an edit takes. Expect hours, not minutes; no beta test run has finished a real recording yet. See [A real recording](https://eddystudio.ai/docs/real-edit/). > **Unknown:** How much the optional vision model improves scene labels. The vision model needs macOS 14 or later, so on older Macs scene labels come from on-screen text and screen movement only. > **Unknown:** Whether you can switch captions off with one setting. You can ask in chat. > **Unknown:** Whether YouTube upload from the Export sheet works in the beta build. The app has a state for "not set up yet". > **Not final:** Eddy's built-in house style (scene layouts, motion and timing). It is still being tuned and may change during the beta. > **Not final:** The Clean, Classic and Bold caption styles. > **Not yet:** A house style learned from your own videos. > **Not yet:** A public download. The first 10 beta creators are hand-picked and free; after that Eddy is a one-time purchase: $49, rising $10 every 5 buyers. ## Getting help Email [support@eddystudio.ai](mailto:support@eddystudio.ai), or use **Send a report to Eddy** in the Doctor sheet. ## Ready to try it? The first 10 beta creators are hand-picked and free. Check the fit first; the link to apply is right after it. ## 30-second fit check Three questions before you apply. Nothing is saved or sent. 1. Is your Mac Apple silicon (M1 or later), on macOS 13 or later, with 16 GB of memory? Yes: Good. Eddy runs on that Mac and your footage stays there. No: Eddy won't run on an Intel Mac or on macOS older than 13. With less memory it warns you that edits may be slow or stop. [See the full list](https://eddystudio.ai/docs/get-started/#what-you-need). 2. Do you pay for ChatGPT or Claude? Yes: Good. Eddy edits through that plan. You sign in once, in your browser; Eddy never uses an API key. No: Eddy needs one of them to edit. Either one is enough. 3. Are you OK with an edit taking hours, not minutes? Yes: Good. Eddy keeps your Mac awake while it works. How long an edit takes is not measured yet, so plan for hours. No: Eddy isn't for fast turnarounds yet. [See how an edit works](https://eddystudio.ai/docs/how-it-works/#how-long-it-takes). [Apply for the beta](https://eddystudio.ai/#apply) --- # Glossary > What each Eddy word means, such as stage, Short, Sub-editor, house style and format, in plain English. Source: https://eddystudio.ai/docs/glossary/ · Written from the app's source code, last checked 3 November 2026; generated tables are rebuilt from the source on every build; not every behaviour hand-tested in the beta build Each word here is used somewhere else in these docs. The names in bold are the ones you'll see in the app. ## Your video ### Recording The raw video you give Eddy: a camera file, a screen recording, or a folder with both. You can also record inside Eddy. Eddy never changes your original files. ### Transcript Every word you said, with the time each word starts and ends. The speech models on your Mac write it. Cuts, captions and Shorts are all made from it. ### Scene A stretch of the video with one layout, such as your face filling the frame, or your screen with you beside it. Eddy labels each scene from what is on screen, and the label decides the layout. ### Stage One step of the progress bar while Eddy edits. The stages are Import, Rough cut, Edit, Render, Check and Deliver. [How an edit works](https://eddystudio.ai/docs/how-it-works/#the-stages) says what happens in each one. ### Long video The main edit of your recording. Its file name ends in `-long.mp4`. ### Short A vertical clip cut from your recording. One edit makes up to 3 Shorts, named `-short-1.mp4`, `-short-2.mp4` and so on. ## Changing an edit ### Chat Where you ask for a change in your own words. The AI you signed in with, ChatGPT or Claude, makes the change. ### Editor A chat that isn't tied to part of the video. It can work on the whole video. ### Sub-editor A chat you open on a scene, or on words you selected. It can change only those scenes and the layers in them, plus motion graphics, which are tied to your words rather than to a scene. ### AI Tools One-click edits in the AI Tools bar: Edit for clarity, Remove filler words, Remove retakes and Shorten word gaps. Style all scenes has no button; the chat and agents run it. ### Retake A part you said again because the first try went wrong. **Remove retakes** cuts the false starts, restarts and repeated phrases. ### Undo Changes to your edit (cuts, scenes, layers, motion graphics and the AI Tools) can be undone, and undoing a cut brings the words back. When the editor rewrites its plan for the edit, that isn't in Undo yet. ### Studio Sound Eddy's clean-up of your voice. It is always on and runs as the footage comes in. ## How it looks ### Your look Your colours, lettering and logo. Eddy uses them on every motion graphic it makes. ### Caption style How the captions look. The Captions panel has 4 styles: House, Clean, Classic, Bold. **House** is the default. ### House style How the whole edit looks and moves: its scene layouts, motion and timing. Unless you ask for one of your saved formats, an edit uses Eddy's built-in house style, which is **Not final**. A format you ask for replaces its timing and structure, in your own look. A house style learned from your own videos is **Not yet** in the app. The house style is not the same thing as the **House** caption style. ### Format The timing and structure of a video you like: the pacing of the cuts, the pauses, the layouts, the zooms and the captions. Eddy measures it from a reference you give it and saves it under a name. It never keeps the reference's colours, fonts, words, music or footage. ## Your Mac and agents ### Doctor sheet The app's health check. It is also where you connect agents, turn on remote access, send a support report and download the vision model. ### Speech models The models on your Mac that write the transcript. One comes with the app. The optional fast transcription model (about 5.6 GB) is much quicker on a long recording; it needs macOS 14 or later and downloads once, when you click **Download**. ### Vision model An optional model that lets Eddy look at each scene, still on your Mac. The vision model is about 3.9 GB, needs macOS 14 or later, and only downloads when you click **Download** in the Doctor sheet. ### Creator tools The 15 creator tools an AI agent such as ChatGPT or Claude uses to run Eddy for you. [For agents](https://eddystudio.ai/docs/for-agents/#the-creator-tools) lists them.