Frequently asked questions

What is Shotline, in one line?

A browser studio for AI-assisted video, for any creator: Build for one-off generations, Grid for wiring shots together on a canvas, Director for shot-by-shot scenes, Signal for cutting the timeline (it is at its strongest on music videos, syncing your shots to a song), and a Library that holds all of it.

What is Early Access, and what do you ask in return?

Right now Shotline is free and invite-only, opened to creators finishing V1 alongside us. In exchange we ask for real feedback (the Report button on any error, or just telling us what feels wrong) and a small credit badge on what you publish. It is an experimental beta: expect rough edges, and expect things to change as we learn from how you actually use it.

Do I need to know anything technical to start?

No. The one thing you do need before you can generate anything is a provider key, which is covered next.

Where do I start? What is "Sandbox"?

Sandbox is the project you land in by default. It always exists and can never be deleted, so a generation you make without picking a project first never lands somewhere you cannot find it. Make a named project whenever you are ready to work on something for real; Sandbox stays your scratch space either way.

How do I get my Shotline access key?

Join the Shotline Discord and run the /claim command in the server. If you are on the early access list, your key appears immediately as a reply only you can see. Save it right away: that reply is private and temporary, and nobody, including us, can read the key back to you afterwards. If you are not on the list yet the bot says so in plain words, and nothing else happens. Adding people is still done by hand, one at a time, so if you are waiting on an invite you are waiting on us and not on a queue.

I lost my access key. What now?

Run /reclaim in the Discord server. Your old key stops working the moment the new one exists, and the fresh key appears as a reply only you can see, exactly like /claim. You can do this a few times a day (three rotations per 24 hours), which is plenty for a genuine loss and low enough that nobody can spin your key in circles.

Does rotating my key log me out or break anything?

Yes, and that is the point. Rotating signs out every device you were signed in on, so sign back in with the new key. It also revokes every connection token you created in Settings, so an AI app you set up by pasting a Shotline token stops working until you make a new token there, in the Your AI section. AI apps you connected by signing in (Claude, ChatGPT, Codex) keep working, so there is no need to rotate your key to fix a connection. Nothing else is touched: your projects, Grids, media, provider keys and settings are all exactly where you left them.

Someone saw my key on stream. What do I do?

Run /reclaim right now. The leaked key dies the moment the new one is issued, and every session it opened ends with it. Apps connected by signing in are not affected by a rotation, so also open Settings, in the Your AI section, and disconnect any app you do not recognise. Whenever a rotation happens, the account also gets a note in its in-app inbox saying the key was rotated through Discord, so if you ever see that note and it was not you, tell us at shotline@thirdaistudios.com straight away.

Do I pay Shotline?

Not right now. Early Access is free. Shotline itself does not charge you for generations: it routes your request to an AI provider using your own account and your own key, and that provider bills you directly for the compute. Read that sentence twice, it is the one thing that trips people up most: Shotline is the routing tool, not the thing you pay for the image or video itself.

So what do I actually pay for, and to whom?

You pay your AI provider directly, on your own account with them, at their prices. Shotline never sees or touches that payment. Think of Shotline as the camera and the provider as the film stock: you buy the film separately.

Which provider should I start with?

Any single key gets you generating; none of them is "the default". The honest trade: fal.ai carries the widest range of models and tends to be the steadiest, while others usually win on price. If you only want one key to start, pick by what matters more to you, range or price, and add a second later. Every card in Settings links straight to where you sign up and create the key.

Do I need more than one provider key?

No, but two is where it gets good. Each provider has its own key slot in Settings, and with more than one stored the studio prices every generation across them and the router can pick the cheaper lane per run.

What does "the router" or "Auto" mean?

Auto is the standing default: for each generation it picks among your own funded, working keys, generally the cheapest lane that carries the model. With one key stored, that key simply always runs. You can also pin a specific provider in the nav if you want every generation to go through one key on purpose; every price in the studio follows whichever you chose.

What is the wallet / "on the keys" thing in the corner?

It is a live read of the balance on each provider key you have stored, so you can see at a glance whether you are about to run out mid-session. It costs nothing to check; it is just a balance lookup, not a spend.

Why does fal.ai say "connected" instead of a dollar amount?

fal only shares your balance with keys created under its ADMIN scope. A normal API-scope key runs generations fine, but fal refuses the balance read, so Shotline shows "connected" instead of inventing a number. Want the dollar figure? Create a new key at fal.ai under Dashboard, Keys with the ADMIN scope and paste that into Settings instead. The trade-off: an ADMIN key can also manage your fal account, so it is a bigger credential to store. Both kinds generate identically.

Why do the utility nodes want a fal.ai key?

The Grid's utility nodes, matte, depth, lipsync, reframe and sound design, run on fal.ai. It carries the widest range of these specialist finishing models and is the steadiest of the providers we route to, which is exactly what you want for a pass you are layering onto footage you already made. The trade is price: fal is usually the most expensive lane, so those nodes cost a bit more than a plain generation. If you plan to lean on them, keep a funded fal key in Settings.

Why do some cost numbers say "about"?

Because providers bill differently. Some charge plain dollars; others sell credits, and credits usually get cheaper the more you buy at once. Shotline always converts credits to dollars at the smallest purchase tier, for example what ten dollars buys, so the estimate is honest for the cautious buyer and pleasantly wrong if you bought a bigger pack. A few providers also report the exact charge after each run; where they do, the number shows plain, and where they do not, it stays an estimate.

Are the prices you show always exact?

Not always, and we would rather say so than pretend. Every price is either a provider's own published rate or one measured from the real charges providers report back after a run, and where a provider hands back the exact charge, the number firms up to that. Where it does not, the figure stays a close estimate, and the real cost is always whatever your provider actually bills you. Prices also sharpen over time: the more a model gets run across the studio, the more real charges we have to measure against, so today's estimate becomes tomorrow's confirmed number. Treat a shown price as a reliable guide, not a contract.

What is the spend meter next to it?

Your own rolling 24-hour spend across your keys, shown next to a daily budget you can set yourself in Settings. It is a personal awareness tool, not a hard stop unless you set it as one.

What is the low-balance warning?

A dollar threshold you set in Settings on a provider balance you want to watch. Cross it and you get a notification (at most once a day), so a generation does not fail mid-session because the tank quietly ran dry.

I got a "such-and-such is out of funds" error. What do I do?

Top up that provider's balance directly on their site (the link is right there in Settings), or switch to a different key you have stored, then try again. Shotline did not charge you for the failed attempt.

What is the difference between Build, Grid, and Director?

Build is the plain one-off generator: pick a model, drop in references, run it. Grid is the canvas where you wire shots, references, and tools together as connected nodes. Director is the shot-by-shot scene surface for building a film out one scene at a time, with camera, lens, and film-stock choices baked in.

What is Signal?

The timeline edit room: it lines your shots up against your song's waveform and lyrics so you can cut the film to the music.

What is Casting?

Your reusable library of characters (Cast), places (Sets), objects (Props), and camera-and-look recipes (Looks), so you build a character, a location or an object once and reuse it across Build, Grid, Director and Signal instead of re-describing it every time.

Why did my generation fail?

A handful of causes cover almost every case: the provider is out of funds, your key was rejected (re-enter it in Settings), you are being rate-limited (wait a few seconds), the model is not enabled on your provider plan, the model's own safety filter blocked the prompt or reference, or the provider is briefly overloaded on their end. The error message on screen tells you which one it was and what to do about it. A failed attempt is not charged, with three exceptions. Some providers charge for a run their content filter declined after accepting it: the decline note on screen says whether it was charged, or sends you to that provider's dashboard when Shotline cannot tell. A decline before the provider accepted the run is never charged, and Shotline stops there and asks before sending it to another lane. If a provider says it accepted the run but never sends back a task id, Shotline tells you so on screen and will not retry it on its own; check that provider's dashboard before running it again. Because it may have been charged, it counts toward your daily spend limit at its estimate until it rolls off. And on a rented GPU card, the card time a failed clip used is spent, because the card is billed by the minute.

Can I undo a generation, or get my money back if I do not like the result?

There is no charge-back through Shotline (we never touched the charge; the provider did). Most surfaces let you compare candidates before committing and keep the ones you want, so the usual move is to just try again rather than expect a refund.

Is my work labeled as AI-made, and can I use it for professional or commercial projects?

We treat content provenance as a feature, not a burden. Shotline is being built to carry an invisible, verifiable "made with AI" mark (the C2PA / Content Credentials standard) inside your outputs: it changes nothing about how your video looks, adds no visible watermark or badge, and never stops you editing, exporting, or commercially using your work. The point is legitimacy for real work, so that publishing platforms can read the mark and label your content correctly instead of penalizing unlabeled AI. We are introducing this in stages by media type, beginning with images, so not every export carries a mark yet. The disclosure side is already in place (for example, our assistants tell you when you are talking to an AI). Whatever we mark, the labeling duty where you publish stays yours: you are responsible for any AI-content label your platform or country requires, and you agree not to strip the mark out once it is there (see Terms 8.5 and 8.2).

What kinds of things can I actually generate?

Three kinds: image, video, and audio. Most of the image and video models take reference stills, clips, or audio too, so you can hold a face, a location, or a voice steady across shots instead of generating from a blank prompt every time. Some of the video models bake in synced sound, dialogue, effects, ambience, as part of the same run, rather than leaving you to add it after. There is also a set of specialized tools, background removal, lip-sync, upscaling, and a few others, for editing footage you already made rather than generating something new.

Which models are actually available right now?

There's a wide lineup across image, video, and audio, and new ones ship regularly, so treat any list as a snapshot rather than a fixed one. On video that currently includes Kling, Seedance, WAN, and Hailuo. On stills it includes Nano Banana Pro, Seedream, GPT Image 2, and FLUX. For music, sound effects, and voice there is ElevenLabs and Lyria, among others. Open the model picker in Build to see the exact current lineup.

Does the same model always run through the same provider?

No. A lot of the flagship models are wired into more than one provider at once, for example Nano Banana Pro runs on three of them and Kling 3.0 runs on two, so if one provider is down or pricier that day, the same model can still run somewhere else. Auto already picks the cheapest working lane for you, that part is covered elsewhere in this FAQ. The point here is just that having a model available does not tie you to a single provider for it.

What is the Toolbox?

The Toolbox is a set of small tools bundled into Shotline that run entirely in your browser: a model price catalog, a music visualizer called ViZ, an audio mastering tool, an MP3 exporter, a lyric sheet formatter, and a file metadata checker. All of it is free. None of it calls an AI model, touches a provider key, or spends anything, so it works the moment you're in the app, provider key or not.

What can I do in the model catalog?

It's a price and capability comparison across every model Shotline supports: what each one takes as input, what it costs on each provider at the same baseline (five seconds at 720p for video, one image at its floor quality, and so on), and a chart of how that price has moved over time. A price in cyan is measured from real bills, and a plain price is the provider's own published rate. A price marked as an estimate is our best guess. Browsing it is free and needs no provider key, so you can shop before you commit to one.

What is ViZ?

ViZ is a music visualizer: drop in a song and it renders reactive visuals live in your browser as the track plays, with no model call involved. Add a still image and ViZ+ animates that image to the music instead of a generic visual. Everything happens locally, the song and image never leave your browser, and how long a render can run depends on your device's own power.

Can I master and export a track without paying for a model?

Yes. The mastering tool brings a mix up to a release loudness target, with presets for platforms like Spotify, YouTube, and Apple Music, using a limiter built to protect the low end rather than just flattening every peak, and it runs on your machine: the track is never uploaded. From there the MP3 tool turns a mastered file into something you can send, at a bitrate you choose from 128 to 320 kbps, with tags and cover art carried along. Both are free and need no provider key.

Does the Toolbox help with lyrics or checking a file before I post it?

Yes, two tools cover that. Lyrics takes a Suno-style sheet and reformats it to what DistroKid expects: it never rewrites a word, it flags anything that needs your own judgment call instead, and saved sheets stay in a small library you can reopen later. The file checker reads a file's metadata (camera, software, timestamps, GPS) and, for images and video, whether it carries a Content Credentials (C2PA) mark, all locally in your browser, so you can check your own work before you publish it.

Which nodes are free to use?

A set of nodes run entirely in your browser using WebCodecs and WebGL instead of calling a provider, so they cost nothing and need no spend confirmation: crop, rotate, and resize; extract frame and contact sheet; speed ramp, datamosh, and gif export; the look effects (dither, glitch, gradient map, posterize, ascii, halftone); ViZ and ViZ+, which turn audio or an image into a music-reactive visual; blend, which layers one visual over another; and stitch, which joins clips into one piece. One exception: crop enlarges the result back up if the cropped region comes out under 1080px on the long edge, so it is not unusably small, and that step is a paid generation. It is a switch on the crop node called "enlarge back up after the crop". Turn it off and a crop of any size is free and comes back at exactly the size you framed. Everything else on the list stays free no matter the size.

What does the Stitch node do?

Stitch joins the clips or audio tracks wired into it end to end into one continuous piece. It follows the order they sit on the canvas, left to right, so you build your cut by arranging shots in space rather than on a timeline. It runs free, client-side, no provider call and no spend, which makes it a natural last node once you're happy with your shots.

Are there limits on the free effects?

Yes. Most of them render live in the browser tab itself, so keep that tab open and in the foreground while a job runs: closing it or switching away mid-render will interrupt it. Speed ramp, datamosh, ViZ, ViZ+, and blend are capped at 30 seconds of clip, and speed ramp and datamosh render video only, so audio gets dropped. Gif export is capped at 12 seconds, and a bigger frame size or higher frame rate makes for a much larger file.

Can I control Shotline from my own AI assistant?

Yes. Shotline runs a Model Context Protocol (MCP) server. The easiest way in is to sign in: add the address mcp.shotline.tools/mcp in your AI app, and it opens a Shotline page where you allow access, with no token pasted anywhere. Claude, ChatGPT, Codex and Claude Code all connect that way. ChatGPT's custom apps accept OAuth or no authentication and have no field for an access token, which is why it signs in. ChatGPT chats and Codex connect separately, even inside the ChatGPT desktop app, so add Shotline in each one you use: both can be connected at the same time. Claude Desktop, Claude Code, Codex and your own scripts can also use an access token from Settings, at shotline.tools/mcp. You can disconnect any of them in Settings. Once connected, your assistant is acting on your account with your own credits. Generation is spend-guarded the same way it is in the app, so anything above a small threshold requires an explicit confirmation before it runs.

What can an AI assistant actually do in Shotline over MCP?

It can generate images, video, and audio on your account, browse your assets and projects, and set up new projects and shots. It can also build on a project's Grid directly: placing nodes, wiring connectors between them, and saving, sharing, or applying reusable node templates. Video and audio generations always need an explicit confirmation before they run, since they spend real credits, and everything it creates lands on your Sandbox project's Grid rather than a separate MCP-only space.

How do I connect my AI assistant to Shotline?

The easiest way is to sign in. In your AI app, add the address https://mcp.shotline.tools/mcp (Claude calls this a custom connector, ChatGPT an app): a Shotline page opens that shows which app is asking, you press Allow access, and it is connected with no token to copy. That works in Claude (and the Claude apps and Claude Code signed in to the same account), ChatGPT, Codex, and Claude Code on its own with claude mcp add. In ChatGPT, set it up on chatgpt.com in your browser: turn on Developer mode under Settings, Security and login, then open Plugins, press the plus button and choose Create app (if an upload box opens instead, press Create MCP App). Name it Shotline, paste https://mcp.shotline.tools/mcp as the Server URL, leave Authentication on OAuth, tick I understand and want to continue, and press Create. To use it, pick Shotline under Developer mode in a new chat; the ChatGPT desktop app picks it up after you quit and reopen it. For Codex, sign in to Shotline in your browser first, then run codex mcp add shotline --url https://mcp.shotline.tools/mcp in your terminal and allow access on the page that opens; it waits five minutes for you. ChatGPT chats and Codex connect separately, even inside the ChatGPT desktop app, where the Settings, MCP servers list belongs to Codex and ChatGPT chats use Plugins: add Shotline in each one you use. Both can be connected at the same time, with the same address. If you prefer a token, or your app cannot sign in: in Settings, in the Your AI section, open Connect with a token instead and generate one. It's shown once, so copy it right away. For Claude Code, run the one-line setup command shown there. For Codex with a token, put the token in your shell as SHOTLINE_TOKEN and run the codex mcp add command shown there: Codex takes the NAME of the variable rather than the token itself, which is the usual reason that setup fails. For Claude Desktop, paste the provided config block into your config file and restart the app. A token uses the endpoint shotline.tools/mcp, and you can mint a separate token per client. Every token and every signed-in app is listed in Settings, where you can revoke or disconnect any one of them without touching the others.

Can I rent a GPU through Shotline?

Renting a GPU is a Studio Plus feature, so it is turned on per account rather than on by default. On an account that has it, you connect your own GPU-rental account and key in Settings, and Shotline rents a physical GPU on your behalf to render clips a normal provider does not carry. What runs on that machine is Shotline's own render setup, not a general-purpose computer. It is a separate lane from the regular provider keys.

Do I need my own account to rent a GPU?

Yes. Shotline never rents hardware on your behalf without your own GPU-rental account and key in Settings, the same pattern as a provider key. You fund your own balance directly with the rental provider, and Shotline can never spend past whatever is sitting in that balance.

When would I rent a GPU instead of using a normal provider key?

Mostly for the handful of models and passes no hosted provider carries yet. Outside of those, a normal provider key is simpler and faster to start: renting a card means waiting several minutes for it to boot before your first clip can render.

Roughly what does renting a GPU cost?

You are billed by the rental provider, by the minute, on your own account with them, not by Shotline. A card runs roughly 70 to 90 cents an hour once it is up. A fresh card also pays a one-time cost to download the model's weight files first, and the estimated cost is shown to you before you start.

Why does spinning up a rented card take several minutes?

A rented card starts with nothing on it, so before it can render it boots, downloads the model's weight files, and warms up. Once nothing is left to render, the card shuts itself down after sitting idle, or you can stop it by hand any time.

What is a "project" here?

A folder: a name, its media, optionally a song, optionally a Grid or Director room. Sandbox is the one you always have; everything else you create as you go.

If I delete a project, is the media gone?

No. Deleting a project (a soft delete) removes it from your workspace but the media itself stays, recoverable, in your account's Assets. The Assets view inside the Library is the only place media is actually and permanently destroyed; everywhere else in the app, deleting is reversible.

What is "Clean up N orphans"?

A render that is no longer referenced anywhere in a project (an old candidate, a replaced take) counts as an orphan. Clean up removes those specific ones and frees the storage they were holding. It only ever touches renders nothing in the document still points at.

What happens when I export or back up a project?

Export downloads a package of the project straight to your device. Deleting a project (outside of Sandbox) saves that package automatically first, so you always have a copy before anything is removed from your workspace; a package can be re-imported at any time.

Can I get everything Shotline has on me?

Yes. Settings has an export that downloads one JSON file with your whole account: profile, projects, Grids, Cast, Set and Prop libraries, presets, and generation history. It does not embed the media files themselves; use a project's package export in the Library for the actual files.

What happens if I delete my account?

It is permanent and immediate: profile, projects, media, libraries, presets, keys, and history, all erased, with no undo. Export your data first if you want a copy. This is the one deliberate exception to the soft-delete rule everywhere else in the app.

How do I share a project with someone?

From the Grid's collaborators panel: either mint an invite link or add someone by their handle and pick a role (editor or viewer). Sharing is per-project, and each collaborator generates on their own provider keys, never yours.

Does someone get added the moment I invite them?

No. An invite lands in their notifications as a pending request; they have to accept it before they show up on the roster. Nothing about your Grid is visible to them until they do.

What is the difference between an editor and a viewer?

An editor can work on the Grid; a viewer can look but not change anything.

What happens to my work if I leave a Grid someone else owns?

Your own renders on that Grid are copied into your Library first, and stay free of your storage quota for 30 days before they start counting against it. You lose access to the Grid itself and everything else on it once you leave.

What is a "nudge"?

A free courtesy ping to your collaborators pointing at a specific spot on the Grid (a node in Grid). It never generates anything and never costs you or them anything.

What is the friends list, and how do I add someone?

Friends is a quick-pick list, not a permission. Being friends with someone doesn't share anything or give them access to your work by itself. To add someone, go to Settings and send a request to their exact handle; they have to accept it before you show up as friends to each other, the same way a project invite works. Once you're friends, their name appears as a one-click pick wherever you share or invite someone, so you're not typing a handle every time. Removing a friend later doesn't undo anything you already shared with them.

Can two of us work on the same project at the same time?

Yes, on both the Grid and the Signal timeline. You will see who else has it open: a small avatar for each collaborator, and a glow when they are actively working. On the Grid you also see their live cursor moving around the canvas with their name and color, and each other's selections outlined, so you know what someone is about to touch. On the Signal timeline you see each collaborator's playhead moving as they work and their lanes tinted, so you can cut side by side.

What happens if we both edit the same thing at once?

Mostly you will not, because on the Grid a node someone else has open shows up locked to you, marked with their name, so you can see it is taken before you touch it. Other changes, moving a node, editing a prompt, adding a piece, are sent individually, so working on different parts of the project you will not step on each other. If you both do land on the exact same thing at the same moment, whichever change reaches the server last is the one that sticks, and it shows up on your screen shortly after. A bigger save that collides with someone else's work is refused rather than silently overwriting it: you are asked to reload before it goes through.

If someone I am working with generates media on a shared project, does it end up in my library too?

If you are an editor on the project, yes, by default: anything that surfaces on the Grid or the timeline copies into your own library automatically as it appears, so you end up with your own copy of the project's media, not just a link to it. It is called Auto sync in your Library, it counts against your own storage rather than the Grid owner's, you can turn it off per project, and anything already copied stays yours even if you leave the project later. Viewers don't get this; they can still save individual pieces to their library by hand.

How do we actually start working together on a project?

Share it with them the normal way and wait for them to accept as an editor. Once they open it, presence, live cursors, playheads, and the project's chat switch on automatically, there is nothing separate to turn on. From there it is just the work: build and wire on the Grid, or cut on the Signal timeline, the same as working alone, except now you can see each other doing it and talk it through in the project's own chat instead of switching to another app.

What is the Grid fullness gauge, and what does "heavy Grid" mean?

The Grid fullness gauge is a small meter that appears in the corner of the Grid once a Grid starts filling up. Its percentage is how full the Grid is: its saved size against the save limit, and how many heavy cards it holds. It stays hidden while there is plenty of room, and it only turns red when the Grid is really close to its limit. If your device drops frames on a Grid for a few seconds while you work, a separate line says so; that is about your device, not how full the Grid is. Click it to see what is driving it and what to do.

Why does my Grid turn into flat colored tiles when I zoom out?

That is intentional, not a bug. When you zoom out far on a Grid with a lot of nodes, the cards simplify into flat, color-coded tiles so the whole Grid stays smooth to pan and zoom. Nothing is deleted or changed: your nodes, wires, positions and selection are all exactly where they were. Zoom back in and the full cards return.

Is there a limit to how big a Grid can get?

Yes, though most Grids sit well under it. A Grid's own contents (its nodes, wires and layout) have a size ceiling, separate from your account's media storage, and the load gauge warns you as you get close so it is never a surprise. If a save would push past the limit, Shotline refuses the save rather than corrupting or truncating anything: you see a message that the Grid is too large and needs thinning, and your last good save stays intact. The fix either way: delete nodes you are done with (your renders stay safe, only the node goes) or move part of the Grid to another project.

Does a big Grid actually slow things down, and what can I do about it?

On some devices, yes, and the load gauge tells you when it is actually happening on your machine rather than guessing. A few things help: delete nodes you are finished with (your renders stay safely in Assets, only the node goes), zoom out while you work, and if the Grid is heavy but everything on it is work you still want, move part of it to another project's Grid instead of deleting anything.

How do I split a heavy Grid into two, without losing anything?

Right-click any node and choose "send to another Grid" to place a copy of it on another one of your projects' Grids. Between your own private Grids, media travels as a reference back to where it was rendered, so moving work around costs no extra storage. Into a Grid you have shared, the file is copied onto that Grid so every collaborator can see it. A node with a finished render sends over as just the media; the prompt, model and settings that made it stay behind on the original Grid. This is also the load gauge's own advice once a Grid goes heavy: delete branches you are done with, or split what you still need across two Grids.

Something broke. What do I do?

Use the Report button that appears on the error. It sends the technical context straight to us. For anything that is not a hard error, or anything you just want to flag, email shotline@thirdaistudios.com; Aidan and Saera read everything there.

Is my provider key safe?

It is stored encrypted, and once saved, no part of it except the last few characters is ever shown again in the browser. You can remove a stored key at any time, which turns off generation through that provider until you add a new one.

Can I turn on notifications for things happening while I'm away?

Yes, browser push is opt-in in Settings: nudges, Grid invites, failed generations, and low-balance warnings can ping you at the OS level. The in-app inbox is always the full record either way; push is just an extra ping layered on top.