The Assistant, Sharing & Team

When you hit a problem in the editor, you reach the assistant three ways: the Ask AI tab in the left drawer, the pill button of the same name at the top of the frame editor, or the Fix it button inside a quality warning. It is one continuous conversation about the whole film, with the selected shot in context; you type what you want into the box at the bottom and follow the answer in the transcript. Even if you close the panel, a running job keeps working in the background.

• Ask AI — a tab in the left drawer (next to Characters/Locations/Assets, ✦ icon) and a pill button above the frame editor. Opens the chat panel; the pill opens it bound to the shot you are editing. The drawer icon turns into a spinner while the AI works; answers that arrive while the panel is closed pile up as a blinking "+N" badge on the icon.
• Fixer AI (panel title) — there is one conversation per project, so the panel simply carries the product's name. The pencil icon beside it closes the running conversation and opens a new one.
• Ask anything about the film… (the message box) — where you say what you want, from one shot's defect to the whole film. 2000 characters max; a live counter (e.g. 1960/2000) appears near the limit and oversized pastes are trimmed. Enter sends, Shift+Enter adds a line break; sending is disabled while an answer is streaming.
• Send — the arrow in the box's bottom-right corner. Dimmed and unclickable when the box is empty or an answer is streaming; shows a spinner during streaming.
• Thinking… — the status shown under the transcript while the answer is being prepared; a faint "AI working" wave appears behind the panel.
• Read-tool rows (list scenes, scene detail, search scenes, cast/location list, production status, cost estimate, balance, reference board, prompt suggestion) — the AI's information-gathering steps show as faint rows in the transcript; they need no approval and are free. Click a finished row to expand its result in a collapsible box.
• Close — closes the floating panel; a running AI job continues in the background and answers pile up as a badge.
• Fix it — from the quality-warning dialog on a shot card, forwards the failed checks to Fixer verbatim: the panel opens bound to that shot with the findings sent as the first message. If you approve, the shot is regenerated; the warning clears itself after a clean generation.

There is ONE conversation per project, and it runs continuously: fix a shot's mouth sync and then say "now tidy up the ending" — the assistant remembers both, because they happened in the same conversation. You never pick a scope: the whole film is always in context, and whichever scene you select in the timeline joins it automatically.

• Whole film — the first pill above the message box. It is always there and cannot be turned off: the assistant sees the scene map, the cast, the locations and the production status, and reads individual scenes itself when it needs them.
• The selected-scene pill — appears beside it whenever a scene is selected in the timeline. That scene's full detail and its frames go to the assistant, and "this shot" or "the man here" resolve to it. The X on the pill takes the scene out of context (without deselecting it in the timeline); clicking a different card moves the focus there on its own.
• New conversation — the pencil icon in the panel header. Closes the running conversation and opens a blank one. Nothing happens if it is already blank. You will rarely want this: the conversation continuing is exactly what stops the assistant from re-proposing a fix it already watched fail.

The conversation is stored in your browser. Starting a new one discards the previous transcript; the AI's approved edits, however, can still be reverted with the editor's normal undo shortcut.

Fixer never spends money or changes content without telling you. Every tool call that spends or modifies stops as an approval card in the transcript and waits for your decision.

• Approve / Discard — the card shows the operation's name in plain language (e.g. "render the video", "edit the scene"), the AI's reasoning and a price tag. Approve applies it, Discard rejects it (the row remains marked "— rejected"). Approved edits are also pushed onto the editor's undo stack — so ⌘Z can revert them.
• Price tag — the amount comes from the very same calculation the server will charge your wallet with, written as dollars + token count together. Free operations say "Free" (free); open-ended jobs that cannot be priced say the amount depends on how far the work goes. The prefix "at least" (at least) means the amount can rise; "approx." (approximately) means the engine is set to Auto and the price is based on the table default. Multipliers that raise the amount are listed by name: "two named characters — a second pass re-locks every face" (×2), "covered hair — a second pass fixes the headscarf" (×2), "includes the thumbnail and face-lock every model page produces".

Free operations such as text and cast patches show as faint rows instead of cards; only generation tools — image/video renders, audio generation, cover art, frame edits, mouth sync and the like — spend money.

If you don't want to approve a stream of small fixes one by one, you can switch the chat to auto-approve. Fixer then applies without asking until the chat's spending cap is reached — but destructive operations are always asked.

• Auto-approve off / Auto-approve on — toggled from the lightning (⚡) menu in the panel header. While on, Fixer applies fixes without asking; when the cap is reached it starts asking again. Deletions and starting production are always asked. Off by default, and kept separately per chat; while on, the lightning button shows the remaining budget in dollars.
• Spend cap for this conversation — inside the lightning menu, the ceiling auto-approve may spend: options of 2,000 / 5,000 / 10,000 / 25,000 tokens listed with dollar equivalents (e.g. 5,000 tokens ≈ $1.43; the default). Below it a ledger line reads "spent: {amount} / {cap}"; when the cap is hit a yellow warning says the limit was reached and Fixer asks again. The cap is also enforced on the server.
• Approved automatically — the mark of a spend nobody clicked: an auto-approved call shows no card; instead the running/finished row carries a lightning bolt + amount. The transcript thus doubles as the chat's spending record.

The safety rules are strict: the cap lives on the server and is clamped to a 25,000-token ceiling whatever you request; unpriceable operations are never auto-approved; deletions (scene/entity/audio clip), reordering the timeline, shortening the film and starting production are always asked; at most 12 operations are auto-approved per turn.

The fastest way to show the film is the watch link. The Share button in the top bar is visible only to the project owner; the dialog it opens manages visibility and the password. The link copied to your clipboard goes to the read-only /watch page, not the editor.

• Share — the share icon in the editor's top bar; opens the share dialog. Owner-only.
• Public link — the switch that opens/closes the film to anyone with the link. It works optimistically; if the server rejects, the switch flips back with a "could not be saved, try again" warning. While on, the link row, the password field and this note appear below: anyone opening the link can leave a note pinned to the exact second of the film; notes land in your email and are listed on the watch page.
• Copy link — the copy icon at the right of the link row puts the watch link on your clipboard; it briefly turns into a green check.
• Password — puts a password on the link while it is public. The empty field reads "set a password"; with one set it shows "•••••• (set)". Apply saves the typed password (unclickable while empty); leaving the field empty when a password is set reveals a red Clear button that strips it.

Whoever opens the link lands on a Netflix-style, watch-only page: the film large, controls below. While playing, the controls fade after 3 seconds of stillness and come back on mouse movement. Only public films (and the film's own team) can be watched; otherwise the page 404s. Right-click is blocked and there is no download button. If the film isn't ready yet, the page says it is not ready to watch.

• This film is password-protected. / Password / Unlock — the screen a visitor sees on a password-protected link: they type the password and press Unlock; a wrong one shows "Wrong password." (wrong password). The owner and invited members never see this screen — they watch directly.
• Back arrow (←) — top-left corner, returns to the runime home page.
• Episode menu — only on multi-episode series: clicking the film title in the top bar opens the episode list ("Episode 1 · Name"); the chosen episode loads on the same page. A single-episode film shows just the title.
• Play / Pause — in the bottom control bar, starts and stops playback; at the end of the film the icon turns into a replay arrow that restarts from the beginning. Next to it sits an elapsed / total time counter.
• Time bar — the slider above the controls scrubs back and forth. For team members, every note shows as a notch on it: yellow = open, white = completed; hovering a notch shows the note's text as a tooltip.
• Fullscreen — bottom-right, enters/exits fullscreen; the icon flips between enlarge and shrink.

The watch page's strongest feature is notes: a viewer pins a note to the exact second of the film, and the team sees and resolves those notes on the same page. Leaving a note requires being signed in — an anonymous visitor can only watch. A submitted note also lands in the owner's email and notification bell.

• Notes / Leave a note — the button in the bottom control bar opens the note panel; the film pauses as it opens and the current second is marked. Team members see the Notes heading with the open-note count (e.g. "3 open"); a signed-in viewer arriving by link sees only Leave a note. A signed-out visitor never sees the button at all.
• Note panel — slides in from the right; notes are listed on a timeline with their minute stamps, and clicking a timecode seeks the film there and pauses. Only the team (Commenter and up) can read the notes; a visitor writes their own note but cannot see the list. Anonymous senders show as "Anonymous viewer" (unnamed viewer); an empty list reads "No notes yet." (no notes yet).
• Note composer (What did you notice?) — opens at the bottom of the panel titled "note at {minute}"; 1000 characters max. The send arrow (or Cmd/Ctrl+Enter) submits; on success it confirms the film's team will see it. On failure it asks you to retry; sending too many notes in a row triggers a "too many notes in a short time, try again in a few minutes" warning.
• Mark as done / Reopen — the check icon next to each note (only for people with edit rights) closes or reopens the note; a closed note's notch turns white on the time bar. If the server rejects, the state reverts.

If you want the film on your own site or a blog, the watch page hands you a YouTube-style iframe snippet. The embedded player is deliberately minimal.

• Embed — the code icon in the watch page's bottom control bar copies the iframe embed code to your clipboard; the button briefly reads "Copied" (copied). It is available to anyone with the link. The embedded page plays only public, password-free films; otherwise the visitor sees a quiet "watch on runime" card.
• Embed player (title + runime logo) — the title and logo in the /embed page's top bar are links opening the film's own watch page in a new tab. The embed deliberately lacks: leaving notes, the episode menu and the back button — only playback, scrubbing, time and fullscreen exist.

/embed checks no sign-in at all: private, password-protected or deleted films never 404 — so no broken box appears on a third-party site — they fall back to the same quiet runime card.

To make the film together, you send email invites from the Team dialog in the top bar. Sign-in is via Google, so you need to invite a Gmail/Workspace address (the dialog says so explicitly). On team projects every generation is charged to the project owner's balance.

• Team — the people icon in the top bar opens the collaboration dialog: email invites, the member/role list, pending invites, leaving the project. The green badge on the icon shows how many people (members + guests) are currently online in the project. The owner invites/removes and manages roles; a member sees the team and can leave.
• Invite — sends an invite to the address typed into the email field ("name@gmail.com"); on success it confirms the invite was created and the user will see it at sign-in. Errors are stated in one line: invalid email, already on the team, invite already pending, member/invite limit reached.
• Role — the permission of an invite or member: Editor — edits the film and spends from your balance; Reviewer — watches and leaves notes, cannot edit or spend; Viewer — watch only. The default is Editor. The owner can change an existing member's role instantly from the menu; if the server rejects, it reverts.
• Remove — the person-minus icon on a member's row (owner only) removes the member. It confirms first: "remove member?" — "{name} will be removed from the project and lose access."
• Leave project — as a member, the exit icon on your own row removes you. It confirms: "leaving the project?" — "your access will close; the owner can invite you again." On confirm the dialog closes and you return to the home page.
• Pending invites / Undo — the bottom section lists invites not yet accepted; the owner can cancel one with the X (Undo). With an empty team a hint explains that the invitee will see the invite on the home screen when they sign in to runime with that email.
• Guests — under its own heading, shows anonymous people currently watching via the share link, live ("Guest" + a green "Online" online dot); with nobody watching it says no guests are watching right now. The same green dot on member avatars shows who is online.

Invites, notes and account events land in the notification bell on the home page.

• Notifications — the bell in the home page's top bar opens the panel; the unread count sits as a badge on its corner ("9+" above nine). Two kinds of content are listed: actionable project invites on top, general events below (export ready/failed, low balance, subscription events, a note left on a shared film, someone joining a project, and so on). Clicking a notification marks it read and takes you to the relevant page; with none it reads "No new notifications." (no new notifications).
• Mark all as read — in the panel's header row, marks all unread general notifications as read; shown only while unread ones exist.
• Accept / Decline — the invite cards at the top of the panel ("{name} invited you to \"{project}\"") accept or decline the invite. On accept your film list refreshes and you are routed straight to the page matching your role: editors to the editor, commenters and viewers to the watch page.

The only paid operations on this surface are Fixer's generation tools; sharing, watching, notes, team and notifications are free.

• In Fixer only write/generation tools (image-video renders, audio generation, cover art, frame edits, mouth sync, etc.) spend money; the price is shown on every approval card as dollars + tokens, identical to what the server will deduct (1 token ≈ $0.0003). Read tools and text/cast patches are free and pass as faint rows instead of cards.
• In auto-approve mode the price shows as ⚡ + amount next to the running row — the transcript is the spending record. The cap is enforced server-side and clamped to 25,000 tokens; deletions, reordering, shortening the film and starting production are still asked every time.
• On team projects every generation is charged to the project owner's balance; only the Editor role can start production. If the owner's balance runs out, a member sees a "the project owner's token balance is insufficient" warning.
• Fixer's limits: 2000 characters per message (the server trims too) and 20 messages per minute per user — beyond that a "too many messages, wait a bit and retry" notice appears. Fixer stays on-topic; it politely refuses attempts to extract system instructions or "forget previous instructions" tricks: "I can't help with that; let's get back to editing the film."
• Deleting a chat cannot be undone (at most 40 chats are kept per project); but the AI's approved edits can be reverted with the editor's undo shortcut.
• On the watch page only public films (and the team) can watch — otherwise 404; right-click is blocked and there is no download button. Leaving a note requires sign-in, reading notes requires Commenter or above, marking a note done requires edit rights. /embed plays only public, password-free films; in every other case it shows the quiet "watch on runime" card.