Product Updates & Releases

What’s new in Anvil

New features, engine improvements, model rollouts, and fixes shipped directly to the Anvil platform and CLI ecosystem.

Current Version

v1.32.0

Last Updated

October 27, 2026

Total Releases

65 Updates

System Status

Operational

1 week ago
v1.32.0MINOR
Latest Release
Added
  • Deck Studio (Pro plan): turn a short brief into a 16:9 slide deck under Decks in the sidebar. Say what the deck is for, who will watch it, how many slides and the tone, then pick one of three design directions and watch the slides build. Revise the whole deck or just one slide in plain words, add an AI image to a slide, read and edit the speaker notes, and go back to any earlier version for free. Export a PDF, a ZIP of 1920×1080 slide images, or a PowerPoint whose text you can click into and edit; exporting costs no credits. You can also ask chat for a presentation and it starts the deck for you, with a card under the reply that opens it in Deck Studio.
  • Edit, comment on, present and share your decks. Turn on Edit to click any text on a slide and retype it, or recolour it from the deck's palette; each save is a free "Edited by you" version. Pin a comment to one element and send it to Anvil, which changes only that element, diagrams included, and replies in the comment thread; a request that needs the whole slide is answered with a pointer to "This slide" and costs nothing. Pick one of your design systems in the new-deck brief or the workspace header, and the deck follows its colours and type. Attach up to five PDF, Markdown, text or image files to a brief so the deck is built from your own material. Present full screen with the arrow keys, a slide counter and optional speaker notes, and share a read-only link that works without signing in, never shows your speaker notes, and stops working the moment you press Stop sharing.
  • Start a deck from just its topic. Anvil asks four quick follow-up questions as clickable options, each with a suggested answer starred: who's watching, what they should decide or do, how to tell it, and how many slides. The questions are free and read the files you attached. If none of the suggested audiences or goals fits, press Generate more for up to three new ones, also free. Pick Answer first for a room that must decide or approve, and the deck leads with the recommendation, gives a one-slide executive summary, proves each reason in its own section, ends on next steps with an owner and a date, and keeps backup slides in an appendix. Before choosing a look, a new Outline step searches the web and lays out the storyline: the main point, two to four reasons and the headline of every slide, with numbered sources you can open. Edit, reorder, add or remove anything in the outline for free; the deck is then built from it, with source lines on the slides, and a fact nobody could find stays a [bracketed] placeholder for you to fill in instead of a guess. Slide text, speaker notes, outline rows and suggested options are written to avoid the patterns that make text sound AI-made, such as "not X but Y" contrasts, one-line closers, forced lists of three and dashes everywhere, in English and Bahasa Indonesia alike.
  • See each deck look before you build it. When you ask for three looks, each card now fills in with that look's real title slide, its colours, its two fonts set in their own type and a one-line description. Open a preview full size and step through the looks, then press Use this look. Previews cost nothing extra, and you can pick and build while they are still drawing; the built deck's first slide follows the preview you picked. A look whose preview couldn't be drawn, and looks made before this change, show as text as before.
  • Make a deck from slides you already wrote. In the new-deck dialog, choose Paste my content and paste a slide script from ChatGPT, Claude or Gemini, up to 60,000 characters. Anvil finds the slides for free and shows each title with its line range so you can check the split, then lays out your exact words in the look you pick: titles, bullets, tables, speaker notes and source lines stay word for word, and [bracketed] placeholders stay for you to fill in. A new Your slides step replaces the outline, so you can read every slide before choosing a look. Decks from pasted content hold up to 40 slides, and long ones build in parts but arrive as one deck in one look. The build price is shown on the button before you press it: the usual deck price for every 10 slides, refunded in full if the build fails. A slide's Visual note is drawn as a simple diagram from that slide's own words, with any label you wrote in quotes. After the build, Anvil fixes small copying slips back to your exact words and warns you about any slide that still differs from your script.
  • Add, move and delete slides after a deck is built. Press + between two slides, or the Add slide tile at the end, and choose Blank slide for a free slide in the deck's style, or Write it with Anvil to describe the slide and have it written in the deck's look for the usual deck price, refunded if it fails. Drag a slide to a new place or use its ⋯ menu to move it earlier, later or delete it. Comments stay with their slide, and deleting a slide tells you how many comments go with it. Every change is its own version, so Restore brings back anything you removed.
  • Work on a deck with other people. Share now has two tabs: View only is the read-only link as before, and Collaborate lets you invite another Anvil account by email and remove them later. People you invite find the deck under Shared with you on the Decks page, need a Pro plan, and can edit slides, comments and AI changes, paying for their own AI with their own credits. Sharing, the brief, the design system, renaming and deleting stay with the owner. Only one AI change runs on a deck at a time. When someone else saves, your workspace updates by itself within a few seconds without touching your unsaved edits, and Versions shows who made each version.
  • New anvil-watcher skill: paste a public Instagram, X, Threads, TikTok, YouTube or Facebook link, or give it a video file, and it watches the video and tells you what happens. You get a short plain summary, then a breakdown with times: the hook, each beat, the cuts and pacing, the words on screen and why it works. Ask for a deep watch and it also reports the motion, music and sound effects. On request it writes a brief that anvil-studio can turn into your own version. It never signs in to any account.
Changed
  • The PowerPoint file anvil-studio makes for a slide deck is now editable. Headlines, body text, lists, footers and buttons are real text boxes you can click into and retype, and cards are shapes you can move and recolor. Photos and charts are separate pictures. Backgrounds, gradients and decoration stay as one picture behind each slide, so the slide still looks like the design. The studio lists the fonts the deck uses; install them first, or PowerPoint swaps in another font and some lines may wrap differently. The PDF is still the exact copy. When the PowerPoint must look exactly like the PDF rather than be editable, ask for the picture version and every slide comes out as an exact image again. If the deck was changed after you approved it, the studio renders it again before exporting, so the PowerPoint always matches the slides you approved.
  • Deep chat now runs on Claude Sonnet 5.5 instead of Opus 5.5, so a Deep answer costs about half as much per word. Deep keeps its own label on replies, so you can still tell Quick and Deep answers apart.
  • Chat answers that search the web cost less. Within one answer, Claude re-reads the earlier search results from a cache, and those re-read words are now billed at a quarter of the normal price instead of in full. A long Deep answer costs roughly 35–45% less; a short answer without a search barely changes.
Fixed
  • A long Deep answer no longer makes the next question summarize the chat early. The chat's size is now measured once per answer, not once for every search round, so earlier messages keep their exact wording until the chat is really long.
3 weeks ago
v1.31.0MINOR
Platform Update
Added
  • Chat Memory now has its own dedicated page under /dashboard/memory with a live capacity allocation meter, category filter tabs (Preferences, Projects, Personal, Rules), keyword search (⌘K), in-place card editing, and safe deletion confirmation.
  • Chat now asks one simple question instead of making you pick a model and a search setting: Quick or Deep. Quick is a conversational assistant for everyday questions. Deep is a researcher that opens with a short direct answer and then goes further, covering how it works, examples, trade-offs, common mistakes and what is still uncertain, in plain language. Both search the web on their own when a question needs current facts, and show sources only when they did. Quick runs on Claude Sonnet 5.5 and Deep on Claude Opus 5.5, now called directly on Anthropic's Claude API.
  • Bring your memories over from ChatGPT, Claude or Gemini. On the Memory page, Import gives you a prompt to paste into your old assistant, then shows its answer as a checklist of facts you can edit, re-categorize or untick, one by one or a whole category at a time, before anything is saved. The prompt asks your old assistant for no more facts than you have room for, and if its answer is still too long, one click trims it to fit, keeping some of every kind of fact rather than cutting whatever came last. Lines that look like instructions to an AI are flagged or blocked, facts you already have are marked, even when your old assistant words them differently, and the import never goes past your 100-memory limit or replaces what you have. It costs no credits.
  • Share a chat answer as an image. The new button under a reply turns the question and its answer, code and diagrams included, into a PNG card in a light or dark theme that you can preview, download or copy. It is made in your browser and costs no credits.
Changed
  • Moved chat memory management from the general Settings page into a dedicated Memory item in the dashboard navigation.
  • The workspace sidebar uses matching icons for New chat, Enhancer, Markdown, Translate and Images.
Fixed
  • The ai-humanizer skill shows up in prompter init again. Its description broke the skill's header, so the CLI skipped it without saying anything.
3 weeks ago
v1.30.0MINOR
Platform Update
Added
  • anvil-studio now shows your real app in videos about it, instead of a flat screenshot of a whole page or an AI-drawn copy. The part of the app the film is about, such as a pricing card or an invoice row, is rebuilt live from your own code and design system, so its labels, values and colours are exactly yours and the video can animate it changing, like "Unpaid" turning into "Paid". When a scene needs proof that the product really ships, the studio takes a sharp picture of just that element from your running app, one per state, using a new studio.mjs capture command. You give the app's address once, and the studio never starts it for you. For a page behind a sign-in it asks you once to save a session and never asks for a password. Next to the directions you see an "App on screen" line (rebuilt live, a real capture, or both), with a recommendation you can change by naming it. Asking the studio to let AI draw your dashboard gets a one-line answer that a drawn app misspells its own labels and shows features it doesn't have, followed by the rebuilt version. "My dashboard" is narrowed to the one piece that tells the story, and AI-drawn mockups are kept for screens you haven't built yet.
Changed
  • The anvil-studio gallery now works like a studio instead of a long report. Your projects sit in a list on the left with a small picture of each, and one project fills the screen at a time with its progress, spend and finals at the top. The work is laid out as large tiles. Click one to see its versions, facts, prompt and file on the right. Click it again for a full-screen view, where the arrow keys move between pieces or slides, V shows another version, C puts two versions side by side, I opens the details and Esc closes. The address bar keeps your place, so a refresh or a bookmark opens the same piece. It is dark by default with a light switch, works on a phone, and is still one file that opens straight from your computer.
3 weeks ago
v1.29.0MINOR
Platform Update
Added
  • anvil-studio now cleans up after a video build. Rendering a film leaves thousands of frame pictures and a second copy of the finished video in the build folder, which added up to gigabytes per project. A new studio.mjs clean command removes exactly those leftovers once the video is in your studio, keeps any video it has not imported yet and says so, and writes one line in the project's ledger with what it freed; --dry-run only lists them. The studio runs it as the last step of every video build, and studio.mjs status shows how much each project could still free, so a project from before this change is cleaned the next time you open it. Your pictures, finished videos, exports and earlier versions are never touched.
  • anvil-studio now lets you decide how a promo or ad film sounds and who is in it, instead of deciding silently. Next to the three directions it lists five choices, each with a recommendation and a reason: a narrator or not, people talking in one scene, faces, words on screen, and music. Say any of them with your pick, such as "A, narator, tanpa wajah". A narrator is chosen from voice samples after you approve the storyboard. A talking scene keeps the clip's own voice, and the studio checks that the spoken words match the script before using it. You can also keep one character in every scene; the studio tells you that a generated face may change a little between clips. Real people, such as the owner of the business, are never generated.
  • anvil-studio now makes a soundtrack for each video with Suno through kie.ai, about 6 cents per track. It is always instrumental, made at the film's length after you approve the storyboard, and comes back as two takes; the studio keeps the one that ends cleanly and runs closest to the film's length. Motion graphics such as a logo sting now offer music too. Say "calmer music" or "faster music" to get a new track, "no music" for silence, or "use my song" for your own. Without a kie.ai key it uses a free catalog track, and with no music source at all it says the film will be silent instead of promising a track.
Changed
  • anvil-studio now shows you a storyboard before it builds any video, not only before paid clips. Promos, motion graphics and ad films all stop after the pictures are made with a storyboard sheet: one card per scene with its picture, its words and a line on the camera move and sound. The gallery shows it with what has been spent and what is still to come. For an ad film, the length question also carries the clip count and the whole film's cost for each option, such as "about 20 s, 4 clips, about $1.50 all in". Words, timing and order stay free to change until you say go; after that, changing a scene means building it again, and for an ad film paying for its clip again. An edit of your own recording gets a sheet only when it adds generated b-roll clips.