What's new

Asking for something like "write quicksort in Java and compare them with a chart" used to classify as a chart request and cap the reply short — truncating the code mid-function. Now a request that clearly wants code gets a generous output budget, so the full code comes through and the chart still renders.

Settings → Usage is now a unified Usage & Cortex hub. A new This chat view breaks down the open conversation’s spend by model, and a routing feed shows how Cortex routed each message — the tier it chose and why. Everything is reconciled to your real billed usage, so the token numbers match across the chat header, this view, and your workspace totals.

And you don’t have to open Settings for the quick version: click the token pill in the chat header and the popover now shows a live Cortex trace for the open conversation — its spend by model and the last few routing decisions — with a link through to the full breakdown.

The Models tab in Settings → Usage now visualizes your usage: a daily bar chart broken down by model (top models + Other, each with its credit total and share), and a usage-by-feature breakdown — so you can see where your credits go over time and across features at a glance, with the detailed tables still there underneath.

When a chat creates a document, its title — shown in Canvas, used for the download filename, and for the Knowledge Base entry — is now a short, meaningful name drawn from the document’s own content (its heading, or a brief AI-generated summary when there isn’t one), instead of a truncated copy of your request.

Titles and filenames are also handled correctly for every language now — a Hindi or Arabic document keeps its real title in the filename instead of being stripped to “document”.

The Cortex Autopilot flow in Settings → Usage — and the “Saved by Autopilot” savings — now light up properly, showing how your requests fanned out across the Saver, Balanced, and Premium tiers and the credits that smart routing saved you. Each request is tagged with the model tier that served it, and the breakdown fills in for your past usage too, not just new activity.

Exporting a Canvas document to PDF now goes through the same server renderer the rest of the app uses, so Hindi, Urdu, Arabic — or any script — come out crisply shaped instead of garbled, and equations are typeset as real math. Your page setup carries through too: page size and orientation, margins, running headers & footers, page numbers, and per-block text alignment and image sizing all land in the PDF as you set them.

This retires the old in-browser PDF engine that couldn’t shape non-Latin text — the last place a document could come out garbled is now fixed.

Settings → Usage has been rebuilt around a story, not a wall of numbers. The Overview tab opens on a spotlight of your total tokens, cost, and credits, then plots a GitHub-style activity heatmap of your daily rhythm — with active-day and streak stats — and an animated view of how Cortex Autopilot routed your requests across the Saver, Balanced, and Premium tiers, including the credits that smart routing saved you.

Everything else folds into tidy tabs: Models breaks usage down per model and per feature, and workspace owners and admins get a per-member breakdown under Members. Range filtering and one-click Excel export are unchanged.

Equations render properly in your generated PDFs now — inline like E=mc² and full display formulas (sums, integrals, fractions) come out as real, crisply typeset math, not raw text. And when you export a themed document, the PDF picks up your theme's colors (headings, accents) so it matches what you see on screen.

Generate a PDF in Hindi, Urdu, Arabic — or any script — and it now comes out crisp and correctly shaped, not garbled. Documents render through a proper typesetting engine with full right-to-left and complex-script support, and the file contains only your finished result — no stray prompt text, and a clean, sensible filename.

Every panel you run now shows up in Panel Library → Run history, with its cost and a one-click re-run — on Pro, not just Team. The rule is simple: if you can run a panel, you can see it in your history.

Every finished AI Panel now shows what that run cost — a small credit chip sits right in the result header next to the expert count and run time, so there’s no guessing and no digging through history.

The Deep Research panel now does real legwork before your experts weigh in. It breaks your question into several focused searches, pulls the freshest results, and reads the full articles — not just the one-line snippets. So when you ask about something current — “who’s going to win?”, “what’s the latest on…?” — it answers from what’s actually happening now, not last year’s training data.

If a reviewer spots a gap mid-analysis, it goes back for one more round of researchto fill it. And you can see exactly what it did — a live trace shows how many queries it searched, how many sources it read, and whether it re-searched.

Because it does more real work, a Deep Research run costs a little more and counts against a monthly Deep Research allowance on your plan — the panel card shows the adjusted estimate before you start.

Made something worth sharing? Hit Share under any answer to create a private link to that conversation. Anyone you send it to opens a clean, read-only copy — no account needed.

It’s a snapshot, so anything you say afterwards stays private, and you can revokethe link anytime. Only the exact chat you shared is ever visible — nothing else in your workspace.

Want tighter control? Add a password or set the link to expire (1, 7, or 30 days). And the images and audio in a shared chat now come along for the ride — they stay put even after you keep working.

Ask for a rich piece — say, “write the first page of a mystery, add images along the way, and end with a spoken narration” — and AetherIA now composes it as one seamless response: flowing text, images placed at the right moments, and a closing audio piece you can play.

While the narration is being composed you’ll see a small animated indicator, so you’re never left wondering.

The Share this chat window now matches the rest of Aetheria — proper panel, aligned fields, and a gentle fade-and-scale as it opens. Press Esc to close it, just like every other dialog.

The same subtle motion and hover polish now carry across dialogs and menus app-wide, so pop-ups feel of a piece in every theme.

Asking for spoken audio now works however you phrase it. Before, “generate the audio” or “make an audio piece” could slip past and come back as plain text — or, worse, a reply saying we couldn’t make audio at all. Now those route straight to the real audio pipeline and narrate it for you.

And AetherIA no longer claims it can’t generate audio (or images, or diagrams) or point you to an outside tool — it makes them itself. Just ask, e.g. “generate the audio of this.”

When you chose Standard while turning a chat into a document, or added certain ready-made experts to an AI Panel, the request could quietly fall back to a default model instead of the balanced one you picked. That’s fixed — Standard now routes to the intended model every time.

We also added a build-time check so this whole class of routing mix-up can’t slip back in.

Longer audio — like a full news summary — now generates reliably. Behind the scenes we narrate it in short segments and stitch them together, so a single long request can’t stall or time out.

And a small nicety: if you slip and type “adio” instead of “audio”, we still know you want audio and route it to the right place.

Ask chat for an audio clip about the latest news, today’s scores, or anything current, and it now pulls in live web results first — so it narrates real, up-to-date headlines instead of saying it doesn’t have access to current information.

Only time-sensitive requests trigger the lookup; evergreen topics (explainers, stories, overviews) are authored exactly as before.

Audio you generate in chat now comes with topic-related cover art and a cleaner, produced-feeling player — play/pause, a smooth scrubber, and a running time. Want it out of the way? Hit the open-in-a-new-tab button to pop the audio into its own browser tab (where you can also save it).

And every generated audio now ends with a short spoken “generated by AetherIA — double-check anything important” note, with a matching line under the player, so it’s always clear what’s AI-made.

Ask chat to “make an audio overview of…” or “explain this as audio” and you now get a real spoken overview — not your request read back to you. We turn your ask into a natural narration script first, then speak it, so it sounds like a briefing rather than a robot reading an instruction.

And when something genuinely can’t be turned into audio, you’ll get a clear, friendly message telling you what to try instead — never a cryptic error.

Hit Present on a slide deck and it docks to a side rail — the rest of the app slides over to make room, so you can reference your chat, notes, or Knowledge Base while you present, and move slides right there. Leave the Lyceum and the presenter comes with you.

Use ← / → to move between slides from anywhere, toggle speaker notes, and press Esc to exit. The sidebar tucks away automatically so the deck gets center stage.

Your video overviews now keep playing as you move around the app, just like audio. Leave the Lyceum and the video shrinks into a floating player that keeps going; come back and it settles right into the page.

Want it out of the way entirely? Hit Picture-in-Picture to pop the video into its own always-on-top window while you work anywhere — in the app or another tab.

Press play on an audio overview and it now follows you. Head anywhere else in the app — chat, your Knowledge Base, settings — and a floating mini-player keeps the audio going, with play/pause and a scrubber right there.

Drag it to whichever corner you like — it remembers the spot and quietly steps aside when a panel needs the room. Your device's media keys and lock-screen controls work too. Return to the overview and the player settles right back into place.

Generate an audio overview (briefing or podcast) or a video overview in the language you study in. Open the customization sheet (the › on any overview) and pick a Language: English, Hindi, Urdu, Spanish, French, German, Japanese, Chinese, or Korean.

The narration is written and spoken in your chosen language — same natural voices, grounded in your own sources. English stays the default, so nothing changes unless you pick another language.

The Live Search control in Settings → Intelligence is now a clear three-way choice: Auto (search the web only when your question needs it), On (search every message — the LIVE SEARCH badge shows in the composer), or Off (never search).

Previously a single toggle was mislabeled and search was effectively forced on for everyone, so turning it off didn’t stick. Now Auto is the default, and whatever you pick is respected.

Press / at the start of the message box to open a quick-command menu — just like the tools you already know. Filter as you type, arrow-keys to move, Enter to pick.

It drives the controls you already have: /search toggles Live Search, /council opens multi-model deliberation, /fast · /balanced · /reasoning set the answer speed for your next message, and /attach, /new, /settings are a keystroke away. A faster path for power users — nothing changes if you never press “/”.

If you’re on an individual plan, Settings now calls your space a Workspace (not a “Team”), with a clean preview of what a team unlocks — roles, a shared knowledge base, an audit log — for when you’re ready to invite people. The full team-management tools stay where they belong: on the Team plan.

We also moved “Require two-factor on this account” into Settings → Security for individual accounts, so your personal security controls all live in one place. (On Team plans, org-wide 2FA enforcement stays under Team.)

We made two-factor authentication much harder to get stuck behind. Settings → Security now warns you when your recovery codes run low and nudges you to add a backup method (a second passkey or an authenticator) if you only have one — because a backup beats a recovery every time. The sign-in screen also makes the “Lost your device? Use a recovery code” option clear and obvious.

And if you’re truly stuck, a workspace owner or admin can reset a member’s 2FA from Settings → Team (you set it up again on your next sign-in, and you get an email about it) — and our support team can do the same for individual accounts.

The Study Plan tab now shows a study activity map — a full year of your reviewing as a grid of day-squares, shaded from light to deep by how much you studied each day, with your current streak right beside it. It's the long view a single number can't give you: your consistent stretches, the gaps, and the momentum building back.

Hover any day to see the exact count. It fills in automatically as you review your flashcards — nothing to set up.

We hardened how you sign in and stay signed in. Sign-in now uses a more secure exchange that keeps your access token out of the web address entirely — so it can't end up in browser history or logs. You won't notice a difference signing in; it's just safer underneath.

You can now see every device and session signed in to your account under Settings → Security, and sign any of them out — or sign out everywhere — in one click. When you do, access is cut off immediately, not minutes later. And sensitive actions (like deleting your account or changing billing) now ask you to confirm your identity first, so a left-open tab can't be used to make them.

Add-on credits are no longer limited to a few fixed packs. Pick any amount with a slider or by typing it — from a small top-up to a large one — and see the price update live before checkout. A few quick presets are there too. Credits never expire and stack on your monthly allotment.

Deleting your account no longer wipes everything instantly. Your account is now scheduled for deletion with a recovery window — if you change your mind (or it wasn't you), you can recover it with one click and your data is restored. After the window closes, everything is permanently removed, and we retain only the minimal records the law requires (e.g. billing for tax). Safer, and fully under your control.

Download My Data and Delete My Account (Settings → Legal) now work end-to-end. Your file-upload consent follows you across devices and appears in your data export, each policy shows whether you've accepted it (and when), and the Settings search now filters as you type.

Under the hood we added an app-wide safeguard so AetherIA never shows technical error details. If something goes wrong you'll always see a clear, friendly message — never internal system text.

When you ask an AI Panel something time-sensitive — "who's winning the World Cup?", "the latest on X" — the panel now searches the live web first and grounds every expert's analysis in current data, instead of answering from memory. The result shows a "Live data" badge and the sources it used, so you can check them yourself.

When you build your own AI Panel, each expert now picks an intelligence level — Fast, Balanced, or Reasoning (with an optional Thinking boost) — instead of a long list of raw model names. AetherIA routes each pick to the best current model for you, so your saved panels keep working as models improve. Cleaner to set up, and consistent with how the chat composer already works.

When you reach a feature on a higher plan (like AI Council or the spaced-repetition review queue), you now get an animated preview of what it does plus a clear, personalized list of everything that plan unlocks for you — generated from your current plan, so it’s always accurate. One click takes you straight to the right plan, pre-selected.

Saving a generated slide deck to Google Drive or OneDrive (as PowerPoint) used to drop the slide images and re-style the text generically. The PowerPoint is now built directly from the deck — so each slide's image is embedded and the layout matches what you see in AetherIA. (The local "Download PDF" was already correct.)

A generated Video Overview in Lyceum used to show a plain black frame until you pressed play. It now opens on a branded, topic-aware AetherIA cover — the same title card the video itself starts with — so the player looks finished at a glance and in shared links.

Once in a while a reply could stall for ~30 seconds when a quick background step (like naming the chat) hit a slow moment with one AI provider. Those steps now fail over to a backup in a few seconds instead of waiting it out, and the chat title no longer holds up your answer — so replies start streaming right away.

A tall generated image (like a portrait poster or a newspaper front page) used to overflow the chat and push everything else off-screen. Images now scale to fit the conversation, with the full-size view still one click away via the expand button.

When you ask for more than one kind of output in a single message, AetherIA now reliably delivers all of it instead of bailing with "I'm a text-based assistant, I can't generate images."

• "Generate a cartoon of today's top headlines" — pulls the live news, then draws it.
• "Diagram the flow and write a summary" — a real diagram and a full explanation.
• "Mind map photosynthesis and a hero image of a leaf" — both, in one reply.

And when live data isn't available, it tells you so plainly rather than inventing confident-sounding facts.

Ask for "a map of India" (or any city, region, or landmark) and AetherIA now drops a real interactive map right into the conversation — pan, zoom, and a pin on the place — instead of sending you off to another tab.

Maps render from genuine map data, never an AI-drawn picture, so the geography is accurate. Asking for a data map (e.g. "population by state") still gives you a chart.

Mind-map node labels now fit neatly inside their boxes instead of spilling over the edges, and the section colours are a couple of shades deeper for a cleaner, more legible look.

A mind map no longer changes colours when you leave a chat and come back, and its node labels stay crisp instead of clipping. AetherIA now always shows the live, theme-correct rendering rather than swapping in a previously-cached image that was drawn with a different palette and font.

A new signature theme in Settings → Appearance: Aetheriapairs a warm near-black palette with an editorial reading style — AI responses render in a refined serif (Source Serif 4) with crisp Inter headings, so answers read like a polished publication rather than a chat dump.

It's opt-in — pick it when you want it; every other theme is unchanged. Code blocks and diagrams stay in their usual fonts.

AI responses now read like a polished product, not an emoji parade. By default AetherIA keeps decorative emojis out of headings and bullet points while preserving meaningful marks (✓, ✗, ⚠). Code and diagrams are never touched.

You're in control: Settings → Appearance → AI Output Style lets you choose As-is (keep everything), Minimal (the new default), or Clean (no decorative emojis at all).

In a long chat that jumps between topics, earlier subjects no longer bleed into later answers. Ask a fresh question and AetherIA answers that question — it won't drag in a previous topic just because your wording happened to contain a word like "it" or "this".

Genuine follow-ups still work exactly as before — "tell me more", "make it bigger", or "translate that into Spanish" keep their context. Only unrelated, self-contained requests are kept clean.

You no longer have to name a diagram type. Ask for "a visual overview of photosynthesis", "map this out", or "explain the lifecycle of a support ticket" and AetherIA now renders a real, interactive diagram — a mind map, a state diagram, whatever fits — instead of a plain text sketch.

It also draws from the right source: ask to visualize your own notesand it builds the diagram from them; ask about a general topic and it generates one from world knowledge rather than saying it can't find it in your documents.

AetherIA now picks the visual that fits your content instead of defaulting to a flowchart. Ask how something works step-by-step and you'll get a sequence diagram; ask for a lifecycle and you'll get a state diagram; ask for the key ideas of a topic and you'll get a mind map — with charts, ER diagrams, timelines, and more all on the table.

Just as important: when a diagram wouldn't help — a plain explanation, a definition, a set of practice problems — AetherIA now simply gives you a clear written answer instead of bolting on a visual you didn't ask for.

You can now scroll back through a conversation while AetherIA is still replying — the view no longer snaps you to the bottom on every word. It follows the latest text only while you're already at the bottom; the moment you scroll up to re-read something, it stays put, and a “Jump to latest” button brings you back down whenever you want.

Every reply also now carries a small reminder that AetherIA can make mistakes — double-check anything important.

The plan ↔ chat loop now closes both ways. You could already open a chat from a plan task; now a study chat that completed a task shows a “Back to plan” chip at the top — one tap takes you straight back to that plan in the Lyceum, so you never lose your place between studying and planning.

Finished an exam? Archive its plan to clear it from your active list without deleting anything. Archived plans are now fully retrievable: open Archived at the top of the Plan list to see them, and Unarchive any one to restore it. (Delete stays as the permanent option.)

Every task in your study plan now has a chat action: tap it and AetherIA opens a guided chat that teaches that topic from your own notes(or builds the flashcards the task calls for). When you've studied it, the task checks itself off — with an Undo if you disagree — and a View chat link keeps that conversation attached. Your plan's progress reflects both the tasks you've finished and the cards you've actually retained.

Everything also stays organized by course: each plan belongs to a course, the Plan list follows your course switcher, and when you assign a course AetherIA offers to move the chats you studied from that plan into it — so a course's notes, generated material, plan, and conversations all live in one place.

At the top of every study plan, a Coach now tells you the single most useful thing to do next — and why. It watches the whole picture (days to your exam, what's due for review, today's topic, whether your notes have flashcards yet, your streak) and turns it into one encouraging, one-tap action: review the cards about to fade, study today's topic from your notes, make a recall deck, or catch up.

It always leads with the highest-impact move (active recall first, grounded in the science of spacing and retrieval practice), it's honest — no fake urgency when you're on track — and it never spends credits without your tap. Included on the Student plan and above.

Your notes now have a home. The Lyceum (the brain icon in the sidebar) is a study workspace organized per course, with four tabs: Create turns your sources into a study guide, flashcards, a quiz, a mind map, a data table, an infographic, a narrated audio overview, a slide deck, or a video — each running in the background, several at once; Library keeps everything you've generated; Plan lays out a day-by-day exam countdown; and Review is a daily spaced-repetition (FSRS) queue that schedules your flashcards so they stick.

Add a deck to Review with one tap, enter an exam date to get a schedule whose checkpoints feed that queue, and watch your progress keep the plan honest as the exam approaches. The Lyceum's generators, review queue, and planner are included on the Student, Pro, and Team plans (the Data Table is free on every plan).

Canvas has a new Study Guide preset. Open Create Your Document and pick Study Guide to synthesize your chat or notes into an exam-ready guide — a clean overview, a list of key terms with definitions, the core concepts, and a set of likely exam questions to test yourself against.

Study Guide is part of our push to make AetherIA the best place to study, and is available on the Student plan and above.

The Knowledge Base is for understanding the material you upload — not for re-implementing a word processor. Imported documents now open as read-only sources you analyze and chat with. AetherIA reads them; it never rewrites them.

When you want to create something, AetherIA synthesizes a new document from your sources — leaving your originals untouched. Datasets and spreadsheets (CSV, TSV, XLSX, ODS) open as a read-only view to summarize and compare, and a non-mutating Diagnose & improve review still flags stale facts, gaps and contradictions in a source.

Every page in the docs now opens with an animated illustration and carries a diagram — a flow, a decision chart, a layered overview, or a comparison — so a concept is something you can see, not just read.

The Diagrams page now shows a live rendered example of each output type (Mermaid, PlantUML, JSON charts, and real KaTeX math), and there's a new visual walkthrough of how your Knowledge Base is kept private and secure.

Every document you add to your Knowledge Base is now encrypted at rest with a key unique to your organization. That key never reaches your browser — documents are encrypted and decrypted only inside our controlled servers.

One organization's key can never unlock another's. Documents indexed before this release were encrypted in place, so the protection covers everything already in your Knowledge Base.

Adding documents used to mean tabs, a queue, and an "Index All" button. Now it's one screen: drop a file — or connect a cloud drive — and it indexes automatically. Duplicates are skipped quietly.

Everything indexed lives in a single library you can search by content, not just by filename, with each document showing how often it has helped answer a question.

"Summarize today's news" now reliably returns real headlines, and asking AetherIA to summarize your own documents draws on their actual content instead of guessing from filenames.

A round of reliability fixes also hardened document retrieval, file uploads, and how attached files are handled within a conversation.

12 new doc pages live — Canvas, web search, diagrams, widgets, subscriptions, tax & invoices, data handling, PII detection, GDPR, account & sign-up, choosing a plan, and common errors. The sidebar now has 20 live pages and zero "coming soon" markers.

Local docs search and a public changelog (this page) ship in the same release, with instant search across every page as you type.

Verifying you're a student is now near-instant. AetherIA recognises academic email domains and checks a few trusted signals automatically, so most students get the Student plan in seconds — often with zero clicks — and only fall back to a quick manual step when needed.

You'll also get friendly reminders ahead of your annual re-verification, with a grace period so access never cuts off unexpectedly.

Pricing is in USD worldwide, with tax (VAT, GST, and sales tax) calculated and applied automatically at checkout. EU and UK businesses can add a VAT ID for reverse-charge treatment. More currencies are on the roadmap.

Students with the GitHub Student Developer Pack get another fast lane to the Student plan — a verified academic email on your GitHub account is recognised automatically.

Three power-user features:

• Canvas version dropdown — browse and restore prior versions of any artifact, with hover-to-preview side-by-side diffs and a 10-second undo banner.
• Regenerate response — re-roll an assistant bubble without losing the original; compare side-by-side.
• Fork conversation — branch a session from any user message into a new sibling session, with optional auto-resend.

A round of polish across the app: a jump-to-latest pill in long chats, copy the whole message in one click, an offline banner when your connection drops, Canvas ZIP export, and a more accessible session-expiry prompt.

Ask for a story with images and AetherIA streams the prose while images render progressively in the right places. Character consistency across images works out of the box — describe the character once, the model holds the design across every panel.

Auto-routing now always reaches for the newest version of each model family, so you automatically benefit as models improve — no settings to touch. Council mode also gets a visual consensus bar: per-member vote chips and a bar that fills as the experts agree, so you can see how strong the consensus is at a glance.

Council mode now spans Claude, Gemini, and any custom mix of experts you pick, so a question can be weighed by several models at once. The compact card shows the final consensus and confidence; click to expand the full transcript and see how each expert voted.

AetherIA now decides when a long response belongs in Canvas (its own editable document) versus staying inline in the chat — and when it's a close call, it simply suggests Canvas rather than forcing it. Longer, document-shaped answers get room to breathe; quick replies stay where they are.

Quality-critical features — diagrams, widgets, tables, and search answers — now run on our highest-quality Claude models, with automatic failover if a provider has a hiccup. The result is more consistent quality, faster responses, and stronger data-handling guarantees, with no change to how you use AetherIA.

A complete subscription experience: plan selector, upgrade and downgrade flows, top-up packs, invoices, payment methods, and a year-end summary. Free, Starter, Pro, and Team tiers all live.