Changelog

Web app

Changelog for the Neutropic web app (neutropic.io).

v0.1.63

October 26, 2026

  • ImprovementYearly billing includes the same monthly tokens as monthly billing: The yearly discount now applies to the price only. Lite, Pro and Max show the same monthly allowance (≈ 12M, 28M and 72M managed tokens) whether you pick Monthly or Yearly on the pricing page, the plan page and in the apps, and the pricing FAQ says so. If you are already on a yearly plan with the smaller allowance introduced earlier this month, your balance is raised to the full monthly allowance.
  • FixSettings → Billing crashed the page: Opening Settings → Billing (or the /home?settings=billing link) as a signed-in user showed "Application error: a client-side exception" instead of the billing panel. The panel loads again.

v0.1.60

October 23, 2026

  • FixFigure → Canva notice no longer prints the long edit URL: Sending a figure to Canva now shows "Uploaded as an asset and opened in a new Canva design." with an Open ↗ link, instead of several hundred characters of URL in the toast.
  • ImprovementCanva post with screenshots: The Canva connector blog post now shows the real screens — the Canva design picker, Download → Canva on a figure, the approval card for creating a design, and the design detail view — in all four languages.

v0.1.59

October 22, 2026

  • FeatureCanva connector is available to everyone: Canva's app review is approved, so the Canva card is back on the connectors page and in Customize → Connectors, the Canva docs page is public, and any Canva account can be connected — not only the team that owned the app during review.
  • ImprovementCanva connector post on the blog: A new Product post walks through sending a figure to Canva (Download → Canva), exporting a deck to Canva, picking a Canva design as an attachment, the default Canva tools with example prompts, and per-tool permissions.

v0.1.58

October 20, 2026

  • FeatureBilling tab in Settings: Settings → Billing shows your current plan with Adjust plan, a cancellation banner with Resubscribe when a cancellation is scheduled, the saved payment method (brand, last four digits, expiry) with Update / Add payment method through Paddle's customer portal, your usage-token balance with Buy more, and an invoice table with a Download PDF per payment in the company statement format. App Store and Google Play subscribers see where to manage their card and receipts instead. The deep link /home?settings=billing opens it directly.
  • FeatureCancel from the Billing tab: Paid Paddle subscribers get a Cancel subscription section at the bottom of the Billing tab with the same end-of-period explanation and inline confirmation as the plan page; once scheduled it shows the end date and Keep subscription. A subscription reserved to start after a plan pass offers Cancel reservation; store subscribers get the store link.
  • ImprovementBilling tab in the iPhone and Android apps: The same panel in the apps, without web-payment links — Manage subscription opens the store's subscription sheet for purchases made on that device, invoices open as an in-app preview, and cancelling a web subscription from the app works too. Arrives with the next app build.
  • Improvement"Top-up" is now "Usage tokens": The plan page, pricing comparison and FAQ, the in-app purchase card, the weekly-limit banner and its Buy usage tokens link use the new name in all four languages (사용량 토큰 · 使用量トークン · 用量额度).
  • FixSign-up with an address that cannot receive mail says so: If the confirmation email cannot be delivered (for example a placeholder domain such as example.com, or a typo in the domain), sign-up, resend-confirmation and forgot-password now show a clear notice — "We couldn't send an email to this address. Check that it's correct and can receive mail, or use a different address." — both inline and as a toast, instead of a generic error. Auth error messages (wrong password, already registered, weak password, expired link…) are now translated into Japanese and Chinese as well as Korean.
  • FixPayment-method update tab: When Paddle's portal link could not be loaded, the pre-opened blank tab is now closed instead of being left open.

v0.1.57

October 19, 2026

  • FixSupport chat on phones: Opening the support chat from a chat with messages on a phone (mobile web or the app) could collapse the panel to its header or show nothing, as if it had opened behind the page. The panel is now a full-screen sheet on phones; tablets and desktop are unchanged.
  • ImprovementClearer error notices: Error messages across the app no longer show server or code text (such as raw JSON from a payment provider). You see a short notice — "Something went wrong. Please try again in a moment, and if it keeps happening, contact the Neutropic team." — in your language, except for cases you can act on yourself (invalid promotion code, rate limits, offline), which keep their specific wording. Login, sign-up and password pages translate the common authentication errors the same way.
  • ImprovementSubscribe from the app while on a plan pass: On the mobile plan screen, an account with a promotion or tax-invoice pass can now subscribe or switch plans: the current plan's card reads Keep {plan}, the others Switch to {plan}, and a notice explains that the store charges today and the subscription replaces the pass from today (remaining pass days are not carried over; the web keeps the pay-when-the-pass-ends reservation). Needs the app build released the same day.

v0.1.56

October 18, 2026

  • FeatureAndroid app on Google Play: The mobile app announcement now covers Android as well as iPhone and iPad: the landing banner links to both App Store and Google Play, the post-login pop-up has two download buttons, and the mobile docs page links to Google Play in its Android section.
  • Improvement"Open in the app" bar on mobile browsers: On an iPhone or Android browser, the landing page and every signed-in screen show a slim "Neutropic official app" bar with one Open button. It opens the installed app, or the App Store / Google Play page if it is not installed. Close hides it for seven days.
  • FixTwo navigation bars on localized landing pages: On /ko, /ja and /zh the sticky header was shown on top of the hero's own header; the localized landing pages now behave like the English one.

v0.1.55

October 16, 2026

  • FeatureNext steps after every answer: Every finished turn — a plain question, a literature review, an analysis, a draft — now ends with a short list of things to do next, built from what this chat has already done and your research profile. Some suggest going deeper on the same work with a more specific request; others carry a skill badge (for example /hypothesis or /experiment-design) and move you to the next stage of the study. Click one to send it; the badge ones run that skill directly. Suggestions you have already seen, or stages this chat has already completed, are not offered again.
  • FeatureHome cards start you with a real prompt: The Home page lists the nine research skills as cards. Each card opens with a concrete, fully written example request — a named population, instruments and design — instead of a blank to fill in, and once Neutropic knows your field and stage (from the welcome survey and your recent chats) the prompts are tailored to you. The prompt is typed into the composer so you can edit it before sending.
  • ImprovementFour-line composer: The message box shows four lines instead of two, so a longer prompt with a skill badge reads without scrolling.
  • FeatureSurvey analysis skill: /survey scores questionnaire and rating-scale data — SUS, NASA-TLX, UEQ, semantic differential, conjoint, open-ended answers — and the new inter_rater method reports percent agreement, Cohen's and Fleiss' κ, Krippendorff's α and ICC. The skills catalog, docs tables and slash menu are updated (77 skills, 45 analysis methods).
  • ImprovementDeleted files stay visible in the chat: When a file from a turn is later removed in Customize → Storage, the turn's file card keeps its name marked as deleted instead of silently dropping it.
  • FixOffline wording: With no network connection a failed send now says "No internet connection — check your network and resend" instead of claiming your tokens are exhausted.
  • FixSearch snippets: Page descriptions on the landing, pricing, use-cases, skills and connectors pages are within the 25–160 characters search engines display, so Bing and Google no longer truncate or reject them.
  • ImprovementLanding and docs wording aligned with the product: Feature descriptions now say exactly what ships — human sciences (not neurosciences), the current skill counts, and connectors listed as they are (Qualtrics and REDCap in the FAQ) — with promises about shared timelines and IRB exports softened to match.
  • ImprovementYour tokens go further: Managed tokens are now counted at each model's actual cost, with no markup. The same turn uses fewer tokens than before — about half on GPT-6 Luna, 40% on Claude Opus 5.5, and 70% on Gemini, Claude Fable and Sonnet. Plan prices and monthly allowances are unchanged, so every plan gets more turns per month; the pricing page's "Basic turns per month" ranges reflect this.
  • FeatureTop-up tokens: On a paid plan you can buy extra managed tokens from the plan page — ≈ 6.5M ($10 / ₩14,900), ≈ 19M ($29 / ₩43,900) or ≈ 100M ($149 / ₩239,000). Top-ups are added to your balance right away, never expire, carry over month to month, are used after your monthly allowance, and are not limited by the weekly cap. Buy as many as you need; the "tokens used up" notice links straight to them.
  • ImprovementWeekly cap replaces the daily cap: A quarter of your monthly allowance is available each week (Monday to Sunday in your time zone) instead of a quarter of it per day. The usage gauge, the Usage panel and the exhausted notice now say "This week · refills Monday". Promotion and top-up tokens stay outside the cap.
  • ImprovementSee the cost before you send: The Agent menu shows, for each effort level, roughly how many tokens the turn will use with the selected model and what share of this week's tokens that is — so a heavy model on Deep is never a surprise.
  • ImprovementYearly billing shows its own allowance: Yearly plans have a lower monthly price and a correspondingly smaller monthly allowance (Lite ≈ 10.5M, Pro ≈ 21M, Max ≈ 50M tokens). The pricing cards, the compare table and the App Store plan list show the allowance for the billing cycle you select. Existing yearly subscribers keep their current allowance until their next yearly renewal.
  • ImprovementPricing notes moved into the FAQ: The small-print line under the plan cards ("prices per user, before tax…") is gone. The pricing FAQ now answers "How are prices quoted?" (per user, USD or KRW, the checkout amount is the final amount including tax, yearly 14–20% off, tokens don't roll over, local models and your own keys are never metered) and "How does checkout work?" (Paddle, instant activation, receipts, coupons via Promotion for 24 hours).
  • FeatureWhere tokens go: Customize → Usage now breaks your usage down by task, project, chat and model for 7, 30 or 90 days, this month or a custom range, with a project filter and a summary line (turns, tokens, plan usage, artifacts, verified citations, 👍/👎). A project row opens its chats and a chat row opens each turn with its model, effort, steps, artifacts, citations checked and your rating. A "How the work splits" donut shows thinking / tools / writing / review.
  • FeatureUsage report and CSV: Open report prints a one-page usage report for the period (by task, totals, projects, top chats, models, tokens per day) in an invoice-style layout you can hand to an advisor or attach to a grant report; Download CSV exports one row per turn. Each project's card menu has Token usage that opens the Usage panel for that project over the last 90 days.
  • ImprovementUsage and Storage are separate settings pages: Customize → Usage is now two sections, Plan limits and Where tokens go, and Storage has its own page right under it with the storage row and a full file list — every upload and artifact with size, chat and project, filters, search, multi-select and delete (example-project files are read-only).
  • ImprovementRenewal confirmation for Korean subscribers: When Paddle needs your confirmation before a discounted subscription renews at the regular price (a Korean e-commerce rule), a banner and a notice on the plan page show the deadline and open Paddle's customer portal to confirm in one click. Without confirmation Paddle cannot renew the subscription.
  • ImprovementKakaoTalk Channel in the footer: The landing and docs footers link to our KakaoTalk Channel next to X, Instagram and YouTube.

v0.1.54

October 15, 2026

  • ImprovementAdvertising cookies now cover the Google Ads tag: With your consent in the cookie notice (or the Settings › General › Data toggle), the marketing pages load the Google Ads tag alongside the Meta Pixel to measure our advertising. Declining or switching the toggle off keeps both tags out of your browser; the Privacy Policy (Section 8) is updated accordingly.
  • FeatureGemini 4 Argon in the model menu: Available on paid plans, replacing Gemini 3.1 Pro. Your saved model choice is unchanged.

v0.1.53

October 14, 2026

  • ImprovementiPhone app announcement and App Store subscription guidance: The landing banner and the post-login popup announce the iPhone app (shown once per account, with App Store screenshots in your language). The plan screens explain that a subscription is managed only where it was started (web or App Store) and how to move between them; the Terms of Service gain an App Store purchases paragraph, and the docs get a "Subscribing on iPhone (App Store)" section.

v0.1.52

October 13, 2026

  • FeatureShare a chat as a read-only link: Click the Share icon in the chat header, or choose Share from the chat's ⋮ menu, to create a link that opens the chat for anyone — the same screen you see, with the conversation and every artifact, but read-only and without an account. The dialog lists active links with view counts, and Revoke stops a link at once. Link previews in messengers show the chat's title. See Share a chat.
  • ImprovementLink expiry by plan: 7-day and 30-day links are available on every plan; links that never expire are available on paid plans.

v0.1.51

October 12, 2026

  • FeatureSlides: Turn a chat's or project's work into a presentation deck — pick Slides in the left panel or type /slides with the talk length and audience. The deck opens in the viewer with its own navigation and a Present button for fullscreen, and reports, figures and charts from the chat are placed on the slides with their sources. Ask for changes slide by slide ("make slide 4 a two-column comparison").
  • FeatureDownload a deck as PDF, PPTX or HTML: The deck's download menu offers the three formats. In the PPTX, text stays editable and diagrams made in the chat are real shapes and connectors, not pictures.
  • ImprovementNeutropic house style: Slides use the Neutropic look — 16:9, a navy cover and closing slide, numbered section kickers, cards and callouts — with five accent themes. Attach a logo and it goes on the cover and in the footer; the deck has no Neutropic mark.
  • ImprovementFree plan: PDF download only: On the free plan a deck downloads as PDF with a faint watermark; PPTX, HTML and the Connected apps exports are available on paid plans, without the watermark.
  • FeatureExport a deck to connected apps: From the download menu send the deck to Google Drive as a Google Slides file, to Dropbox or OneDrive as PPTX, to Canva as a new presentation, or to Notion and Confluence as an outline page.
  • ImprovementSlides docs, guide and example: A new docs page (Run tasks → Slides) and a blog skill guide in four languages, /slides in the skills catalog, and the Example · Research workflow project now ends with a slides turn.
  • FeatureWelcome survey: After sign-up, five short questions — your role, research field, current stage, how you found Neutropic and the features you look forward to — help us set up the product for you. You can skip it, and it is never asked again. Settings → General → Research profile shows your answers and lets you edit them.
  • FixSearch engines index every page: Each page now tells search engines its own address. Until now every page claimed to be a copy of the homepage, so the pricing page, blog posts and docs pages were left out of search results.
  • FeatureKorean, Japanese and Chinese page addresses: The landing, pricing, blog and docs are available at /ko, /ja and /zh addresses, delivered in that language from the server so search engines and link previews show the translated text. Switching language takes you to that language's address, and every page links its translations.
  • FeatureBlog RSS: /rss.xml, plus /ko/rss.xml, /ja/rss.xml and /zh/rss.xml for feed readers and search submissions.

v0.1.50

October 9, 2026

  • FixErrored turns: When a turn ends with a provider error the chat shows "Stopped here" with the reason, also after a refresh.
  • FixReply language: A follow-up in another language is answered in that language.

v0.1.49

October 8, 2026

  • FixNo ad tracking in the apps: The iPhone, Android and desktop apps no longer show the advertising-cookie banner or load the Meta Pixel.
  • FeatureSign in with Apple: You can sign up and sign in with your Apple ID on the website and in the apps, next to Google.
  • ImprovementSign in without leaving the iPhone app: Google and Apple sign-in open in a sheet inside the app and bring you straight back, without the "Open in Neutropic?" prompt.
  • FixLogging out in the app: Logging out of the iPhone app no longer shows a "PKCE code verifier not found" error.
  • ImprovementInsight reports with a cover and a wide layout: The report header now shows a still from your video or your photo, and when you expand the report panel the report uses the full width, one section per row, instead of one narrow strip.
  • ImprovementInsight reports styled as personal assessment reports: Every Insight card now opens with a masthead (logo, report type, issue date) and a result block instead of the sky picture, with numbered sections — closer to a personality-test report than a social card.
  • FixEmotion report counts people correctly: Someone who moves around or briefly leaves the frame is no longer counted as several people, and faces on posters or paused screens are left out.
  • FixFull-body and distant photos: The emotion report now finds faces that are small in the frame. The card notes that their expression numbers are rough.
  • FeatureClaude Haiku 5.5: Anthropic's newest Haiku is now in the model menu on every plan, replacing Haiku 4.5. It follows instructions much better and reads far longer documents, at the same token use. If you had Haiku 4.5 selected, you're moved to Haiku 5.5 automatically.
  • ImprovementHaiku 5.5 in the new models announcement: The landing banner, the popup on Home and the new models blog post now include Claude Haiku 5.5, in all four languages.

v0.1.48

October 5, 2026

  • FeaturePictures inside documents: The assistant now looks at the pictures in an attached Word, PowerPoint, Hangul or OpenDocument file and at the figure pages of a PDF, so you can ask about a chart or a screenshot in a report. Follow-up questions keep seeing them.
  • FeatureMore file formats: Legacy PowerPoint (.ppt), OpenDocument spreadsheets (.ods), MATLAB (.mat) and R data (.rds, .RData) files can be attached and read.
  • FixNo message without its file: The Send button is blocked from the moment you choose a file until it has finished uploading.
  • FixEvery screen in your language: Messages and screens that were fixed in Korean or English — sign up, log in, password reset, chat edit buttons, rename/move/delete errors, the artifacts page, project and share dialogs, consent and tooltips — now follow your language setting in English, Korean, Japanese and Chinese. Insight guest cards and share links are translated into Japanese and Chinese.
  • FixDialogs on mobile Chrome: Pop-up dialogs no longer slide behind the address bar, and their title and close button stay in view.
  • ImprovementUpload guide: The upload docs in four languages cover the new formats and pictures inside documents.

v0.1.47

October 4, 2026

  • FixAttached documents are read: A Hangul (HWP/HWPX), Word or PDF file attached with @ is read by the assistant. Before, it could answer that it was unable to read the document body.
  • ImprovementMore reliable file reading: Old Word (.doc) and PowerPoint (.pptx) files are read, every sheet of an Excel workbook can be used, long documents are read to the end, and Korean, Japanese and Chinese text files load with the right characters. RTF, OpenDocument text (.odt), e-books (.epub), Stata (.dta), legacy Excel (.xls) and Apple Pages/Numbers/Keynote files are read too.
  • FeatureScanned PDFs and photos: Scanned PDFs (receipts, forms) are read page by page as images, large photos are resized, and iPhone HEIC, BMP and TIFF images are converted so the assistant can see them. Images and scanned PDFs you upload without @ are shown to it as well, and it still sees them on your next message.
  • ImprovementPassword-protected files: Locked PDF, Word, Excel, PowerPoint, Hangul and zip files get one clear message — remove the password and attach the file again. The assistant never asks for passwords.
  • FixWait for uploads before sending: Pressing Enter while a large file was still uploading no longer sends the message without the file.
  • ImprovementUpload guide: "Upload data & reference it" and "Uploads fail or quota exceeded" are updated in four languages.

v0.1.46

October 1, 2026

  • FeatureFour more Insight cards: Laugh count, Brand colour, Music mood and Thumbnail first impression, each with a slash command (/insight-laugh, /insight-color, /insight-music, /insight-thumbnail) in four languages.
  • ImprovementInsight chips in groups: With twelve report types, the Insight composer shows four tabs — Reactions, Speaking, Content and Fun — and opens the group of the chip you picked.

v0.1.45

September 30, 2026

  • FeatureHeart rate, voice and posture reports: Insight now makes a heart-rate card from a face video, a voice card from a recording or video, and a posture card from a body shot (/insight-heart, /insight-voice, /insight-posture). Each card downloads as an image, and the Insight docs describe all three.
  • FeaturePresentation rehearsal, review sentiment, ad reaction and A/B comparison: New Insight cards for a rehearsal video (expression, voice and posture in one card), a file or paste of reviews, an ad's reaction scene by scene, and two versions side by side.
  • FeatureShare an Insight card: A Share button next to Download creates a link to the card, with an expiry (never, 7 or 30 days), view counts and revoke. Everyone in the card must have agreed before you share it.
  • FeatureTry Insight without an account: At /try, pick facial expression, voice or posture, drop a file and get a card in the same page. Sign up to keep it in your account.
  • ImprovementFaster pages and previews: Home, the sidebar lists, the Chats, Projects and Artifacts pages and project pages show placeholders while loading and remember your last list, so they appear at once. Artifact previews load faster and are cached.
  • ImprovementPreviews for more files: Videos show their first frame, PDFs their first page, audio a waveform, and spreadsheets, CSV, JSON, notebooks, Word, PowerPoint, Markdown and HTML reports a short preview of their content.

v0.1.44

September 29, 2026

  • FeatureInsight: facial emotion report: A new Insight tool (View All Tools → Insight, or /insight-face) turns a face video or photo into an easy-to-read emotion report card. The card shows the emotion mix, four key numbers, the emotion flow over time, highlight moments with face thumbnails and the reaction to each scene you name. It is written in plain language for marketing and events, not for clinical, hiring or personality assessment.
  • FeatureSeveral people in one report: When more than one face is in view (up to six), everyone is analysed. The card compares them person by person, shows how long they smiled together and has a tab for each person.
  • FeatureExtra items on request: Ask for reaction style, head movement and moments looking away, one emotion in detail (for example smiles), or an expert action-unit appendix.
  • FeatureDownload the report as an image: The download button saves the emotion report card as a PNG, with every person expanded in group reports.
  • ImprovementiPhone photos: HEIC photos can be used for the emotion report.
  • ImprovementInsight examples, docs and guide: The Insight page has three example chats in each language, with report covers. There is a new docs page (Run tasks → Facial emotion report) and a skill guide on the blog, and the skills catalog lists /insight and /insight-face.
  • FeatureClaude Sonnet 5.5: Anthropic's newest Sonnet is now in the model menu for paid plans, replacing Sonnet 5. If you had Sonnet 5 selected, you're moved to Sonnet 5.5 automatically.
  • ImprovementNew models announcement includes Sonnet 5.5: The landing banner, the first-visit popup on Home and the new models blog post now list Claude Sonnet 5.5, in all four languages. The blog screenshots of the model menu were retaken, and the tokens guide now lists Sonnet 5.5 among the standard paid models.
  • FeatureFigma connector: Figma now appears in Settings → Connectors and the connectors guide, with the token scopes to choose.
  • ImprovementFigma guide with screenshots: The Figma connector docs show the connector settings in every language.
  • FeatureSlack connector: Slack now appears in Settings → Connectors and the connectors guide.
  • ImprovementSlack guide with screenshots: The Slack connector docs show the connector settings and the channel picker in every language.
  • FeatureMiro connector: Miro now appears in Settings → Connectors, on the connectors page and in the docs.

v0.1.43

September 28, 2026

  • ImprovementPrivacy policy: support chat AI provider: The privacy policy now names OpenAI as the AI provider that processes support messages when AI support is on.
  • ImprovementClearer steps for connected apps: Steps that use a connected app now say "Reading from Linear" or "Updating Linear" instead of generic file and save labels.
  • ImprovementLinear in the connectors guide: The connectors blog post now covers Linear, with a real example of creating an issue from a chat.
  • ImprovementDropbox and HubSpot guides with screenshots: The Dropbox and HubSpot connector docs now show the settings and file picker in every language, and the HubSpot tool list matches what HubSpot offers today.

v0.1.42

September 27, 2026

  • FeatureMCP guide in the docs: A new help page, available in four languages, walks through connecting Claude, ChatGPT, Claude Code, Cursor and other clients. It covers where to find the server URL, what each permission allows, the available tools, where results appear and how to disconnect.
  • ImprovementMCP in Settings for everyone: The MCP section in Settings now appears for every account.
  • FeatureChat with our support team: A chat button at the bottom right of the website and the app opens Neutropic Support, where you can ask a question without leaving the page. On phones, the button sits in the app's top bar next to the plan button, and the chat opens full screen. Available in English, Korean, Japanese and Chinese.
  • FeatureSupport conversations you can come back to: See your past conversations with their status (in progress or resolved) and a dot for unread replies. Replies from the team appear in real time, and if you're away you also get them by email. Writing to a resolved conversation reopens it.
  • FeatureAttach screenshots and files: Add up to 5 files per message (PNG, JPG, WebP, PDF, TXT or CSV, 20 MB each), with image previews in the chat.
  • FeatureAsk before you sign up: Visitors who aren't signed in can leave their email and a question on the website. Replies arrive by email and in the same browser, and the conversation moves into your account when you later sign in with that email.
  • FeatureNeutropic AI answers first: When turned on, Neutropic AI answers support questions right away from our documentation, in your language, with links to the relevant pages. If it isn't sure, or the question is about your account, billing or a problem that needs checking, it passes the conversation to our team. You can also tap Talk to a person at any time.
  • ImprovementReplies in your language: Our team's replies are translated into the language you wrote in.
  • ImprovementPrivacy policy covers support chat: The privacy policy now lists what support chat keeps (your messages and files, the email you enter if you're not signed in, and the page, browser and app version when you wrote) and how long it's kept. Your research data and chat contents are not included unless you attach them yourself.

v0.1.41

September 26, 2026

  • FeatureConnect AI apps to Neutropic: A new MCP section in Settings shows your server URL and lists every connected app and access key, with its permissions, project and last use. You can disconnect any of them in one click.
  • FeatureApproval screen for AI apps: When Claude, ChatGPT or another app asks to connect, a dedicated approval page shows the app, the account you're signed in with, the permissions requested and an optional single-project limit before you approve or cancel.
  • FeatureAccess keys with ready-made setup: Create an access key with a name, permissions, project and expiry. The new key is shown once, along with copy-ready setup for Claude Code and Cursor.
  • Improvement"via app" badge: Chats started from an AI app show a small badge with that app's logo in the sidebar and on the chats page.

v0.1.40

September 25, 2026

  • FeatureSee how fast each model uses your tokens: The model menu now shows one to three bars next to each model — more bars means the model uses your token allowance faster, so the same plan covers fewer turns.
  • ImprovementPlans show typical research turns: Next to the token allowance, each plan now shows roughly how many Basic research turns it covers per month, from lighter to standard models.
  • ImprovementConsistent full-screen sheets on mobile: On phones, All tools, chat search, the artifacts list, export dialogs and other list and form windows now open full-screen from the bottom, like Settings.
  • FixChats catch up after your phone goes to the background: If you switch apps while a task runs, the chat now resyncs when you come back — progress and the plan card show the finished state instead of staying "in progress".
  • FixClearer plan wording: Plans, messages and help docs now describe your allowance in tokens throughout, state the Free plan's allowance consistently, and explain that how far it goes depends on the model.

v0.1.39

September 24, 2026

  • FeatureNew model announcement: The site and your home screen now introduce GPT-6 Sol, GPT-6 Luna and Claude Opus 5.5 — which plans include each model and where to pick it.
  • FeatureBlog: meet the new models: A new post explains GPT-6 Luna, GPT-6 Sol and Claude Opus 5.5 — what each is good at, which plans include it, and how to switch.

v0.1.38

September 23, 2026

  • FeaturePlan passes: Customers who pay by tax invoice receive a code that activates a paid plan (for example Pro for one month) from the Promotion menu. The plan page shows when the pass ends; registering another code for the same plan extends it.
  • FeatureReserve a subscription after your pass: If you are on a plan pass, choose a plan on the Plans page to register your card now. Nothing is charged until the pass ends; you can cancel the reservation any time before then.
  • FeatureContext ring: A red ring next to the token gauge shows how full the chat's conversation history is. When it fills, earlier turns are summarized automatically and the chat continues right where it is. When a turn reaches its step limit, Continue picks up the remaining steps.
  • ImprovementGoogle Docs, Sheets and Slides tools: Open Google Drive in Settings → Connectors to see its tools grouped by app — Drive, Docs, Sheets and Slides — each with its own permissions.
  • FeatureNew models in the picker: Claude Opus 5.5 replaces Opus 5, and GPT-6 Sol and GPT-6 Luna replace the GPT-5.6 models. If your saved model is no longer offered, the default is used.
  • ImprovementGPT-6 Luna is the Free plan default: If you're on the Free plan and have never picked a model, chats use GPT-6 Luna, so your daily allowance goes much further. A model you picked yourself is kept, and upgrading switches the default back.
  • FeatureHubSpot and Dropbox in Connectors: HubSpot and Dropbox have their own cards and setup guides in Connectors.

v0.1.37

September 22, 2026

  • FeatureNotion connector: Connect Notion under Customize → Connectors. Search and read the pages you share, import pages as Markdown and databases as CSV datasets, and export a manuscript or report to a new Notion page — every tool starts as "Ask each time".
  • FeatureConnected apps in the composer: The + menu groups Google Drive and Notion under one Connected apps row (hover on desktop, tap on touch screens). The Notion picker shows your workspace the way Notion's sidebar does — Private, Shared, Recently viewed and teamspaces, with pages, sub-pages and database rows one level down — and the Google Drive picker shows your files and folders as a tree too.
  • FeatureNotion tool exposure: By default the agent sees the essential Notion tools plus Neutropic's import/export; turn on Expose all Standard tools in the connector's detail view when you need the rest. Refresh tools re-reads Notion's tool list and Change shared pages reopens Notion's sharing screen.
  • FeatureZotero connector: Connect your Zotero library under Customize → Connectors — through Zotero's approval page, or by pasting an API key. Browse My Library, groups, collections and items in the composer picker, import an item's PDF (or its metadata and abstract), and let the agent fetch bibliographies or save a literature review's sources back to a collection. Writing tools stay blocked unless the key allows writing.
  • FeatureGitHub connector: Sign in with GitHub under Customize → Connectors. Browse repositories, directories and files in the composer picker, import data files and scripts into a chat, and let the agent write files, commit and create branches — each tool with your approval; Expose all Standard tools turns on issues, pull requests and the rest.
  • FeatureConnectors page: /connectors is now public and opens with the four workspace connectors — Google Drive, Notion, Zotero and GitHub — each with what the agent can do, how you connect, Connect in the app and Read the guide, followed by the data-source catalog. The landing page's connector buttons link to it.
  • ImprovementExport to Connected apps: The artifact panel's download menu groups the per-connector exports under one Connected apps row (Google Drive, Notion, GitHub) with the real logos, the same way as the composer's attach menu.
  • ImprovementPassword in Settings: Change password… (or Set a password… on Google-only accounts) now e-mails you a password link — the same flow as Forgot password? — with a 60-second resend cooldown, instead of an inline form.
  • ImprovementTerms and Privacy Policy: Updated for every connector — what Google Drive, Notion, Zotero and GitHub can access, that write actions run only after your approval, and which services receive content only when you connect them.
  • FixConnector toggles: Turning a connector on or off (or connecting and disconnecting) in Customize → Connectors updates the composer's Connected apps list immediately, without a page reload.
  • ImprovementCancel at the end of the period: Cancelling a paid plan now takes effect at the end of the current billing period, as the Terms say — you keep the plan and your remaining credits until then, the plan page shows the end date, and Keep subscription withdraws the cancellation. You also receive an e-mail confirming the date.
  • ImprovementCurrent and scheduled plan: While a cancellation or downgrade is pending, the plan page shows "Current plan → From {date}" at the top, marks the current card "until {date}" and outlines the plan that starts next.
  • ImprovementReview naming: The two review tracks are now called Report review and Peer review everywhere — the review tool page toggle, the /peer-review command, the skill list and the docs.
  • ImprovementDocs: The Notion connector page (with screenshots), the Google Drive page, Uploads and Connector authorization describe the current connectors; the peer review page gets its screenshot and the skills tables list all 17 workflow skills.
  • ImprovementLocal AI guidance: Docs recommend a 4B-class model on 8–16 GB machines and 8B–20B-class models from 24 GB, based on real measurements.
  • FixLocal AI settings: Changing the backend, the hybrid toggle or the installed models now applies to open chats on their next turn, not only to new chats.
  • ImprovementStep-limit notice: A turn that reaches its step limit but has already saved results now says so plainly — the saved results are usable and the remaining planned steps were not run — and the plan card marks those steps as skipped instead of leaving one in progress.

v0.1.36

September 21, 2026

  • FeatureLocal models on this device (desktop app): The model menu gains an On this device group. Install a llama.cpp runtime and recommended GGUF models from Customize → Local AI, or point Neutropic at Ollama, LM Studio or any OpenAI-compatible server; local turns are never metered.
  • FeaturePDF export: The download menu of manuscripts, reports and reviews gains PDF next to Markdown, DOCX and LaTeX — printed as laid out in the viewer, A4 with page numbers, figures embedded.
  • FeatureProfile image: Upload a profile picture in Settings → General → Avatar (centre-cropped and resized in the browser); Remove falls back to your Google picture or initials.
  • FeaturePassword reset and change: Forgot password? on the sign-in page sends a reset link; signed-in users can change their password from Settings.
  • FeatureE-mail confirmation: Accounts created with e-mail and password are activated by the confirmation link, with a resend button and clear messages when signing in before confirming. Google sign-in is unchanged.
  • FeatureNotifications settings: Settings → General gains a Notifications section — marketing e-mails (with the consent date), e-mail language and Task alerts on mobile. A public unsubscribe page honours the link in every marketing e-mail.
  • ImprovementGoogle Drive access narrowed: Neutropic now asks only for the drive.file scope — it sees the files and folders you pick in Google's file picker (Choose from Google Drive…) and the files it creates, never the rest of your Drive.
  • ImprovementSign-up form: A Confirm password field, an Agree to all checkbox, a required "I am 14 or older" row and show/hide buttons on every password field.
  • ImprovementPrivacy Policy and consent: The policy names every model provider you can opt into, discloses ad measurement and loads it only after consent; existing accounts see the consent screen once more with a note of what changed.
  • ImprovementWelcome e-mail: Sent in your UI language (English, Korean, Japanese, Chinese).
  • ImprovementConnectors: Disconnecting a connector asks for confirmation and explains what is deleted.
  • FixLanding menus: The Features / Use cases / Resources dropdowns no longer hide behind the Get started button.
  • FixFigure layout: Content is centred under a title that is wider than the diagram.
  • FixExample projects in your language: My Library no longer flashes the English example projects in a non-English UI.

v0.1.35

September 20, 2026

  • FeaturePeer review (/peer-review): A journal-style review of your manuscript — attach the PDF, DOCX or Markdown and name the target journal; five reviewers score it, an editor issues the decision, and you get a prioritized revision guide and recommended journals. The review tool page gets a Report review | Peer review toggle and reviewed documents show a decision badge.
  • FeatureResearch source chips: The literature-review tool page lets you pick Papers, Patents and Web sources with checkbox chips before you ask.
  • ImprovementCharts: Axis labels, titles and legends no longer overlap or spill outside the chart; older charts are re-laid out when displayed.
  • ImprovementProjects and Chats pages: Fully localized in Korean, Japanese and Chinese, with locale-aware timestamps.
  • ImprovementSidebar: Example projects show an Example badge; View All Libraries and View All Chats open the full pages; on phones the sidebar flyouts open as bottom sheets, on larger screens as centred modals.
  • ImprovementCitation cards: Review reports open with [n] chips whose cards show the reviewed document, the session papers and the re-analysis steps.
  • ImprovementProduct introduction: The /intro-of-product deck is up to date with the current product, including a peer-review slide and fresh screenshots in all four languages.

v0.1.34

September 19, 2026

  • FeatureHome and tool pages: /new becomes /home, and eight tool pages (literature review, study design, data analysis, figures, paper writing, hypothesis, statistics, review) start a chat with the tool's skill pre-set — each with example chats to open.
  • FeatureNew left panel: New Chat, search, Home, Customize, the tools (with View All Tools), My Library and Recent Chats.
  • ImprovementAnalysis and statistics reports cite their sources: Sentence-level [n] citations point to the analysis steps and datasets behind each number, with the same source cards as research reports.
  • ImprovementFigures page: A Diagram | Illustration toggle next to the /figure badge.
  • ImprovementManuscript download: DOCX and LaTeX export the manuscript as written with numbered citations; the citation-style and journal-profile selectors are gone.

v0.1.33

September 18, 2026

  • FeatureDrag-and-drop uploads: Drop files anywhere over the chat to attach them.
  • FeatureUploads up to 1 GB: The upload limit rises from 200 MB to 1 GB.
  • ImprovementNew chat greeting: The Neutropic logo and "How can I help with your research?" sit above the composer.
  • ImprovementMobile layout: A top header bar with the sidebar toggle, symbol and plan button, page-specific header actions, slightly smaller typography and a stacked Suggested tasks list on phones.
  • ImprovementUpload messages: Failures are explained in plain language (too large, storage limit reached, empty file, …) instead of error codes.
  • ImprovementCoupon window: A registered discount coupon is valid for 24 hours; the countdown on the plan page shows the remaining time.
  • FixMath in reports: Escaped percent signs and spacing commands inside inline math render correctly.

v0.1.32

September 17, 2026

  • FeatureAny file, any archive: The file picker no longer filters by type; zip, 7z, rar, tar and gz archives are unpacked on request and the agent works on the files inside. Files it cannot read stay attached and @-referenceable.
  • FeatureProject artifacts: Filter a project's files by chat, delete your own files, and see an Uploaded badge on files you uploaded.
  • ImprovementUploads and generated files separated: The per-turn Artifacts created card counts only generated files; uploads appear in their own tile row and under Uploaded data in the files panel, with a remove button.
  • ImprovementAttachments in the composer: Uploaded or imported files appear as @file badges in the composer (each with an X), instead of a permanent chip strip above it.
  • ImprovementReload matches the live turn: Open panel tabs and the active file are remembered per chat, and Deep-mode thinking blocks are restored where they were shown.
  • ImprovementSimpler figure editor: The download menu offers SVG, PNG, PDF and PPTX; the rail keeps Select, Connect, Add node, Direction, Auto layout, Colours and Undo/Redo.
  • ImprovementDocument download: Markdown, DOCX and LaTeX only.
  • FixChat started from a project page: The question bubble no longer waits for a refresh to appear.

v0.1.31

September 16, 2026

  • FeatureDeepSeek and Kimi models: DeepSeek Flash (and V4 Pro) and Kimi K3 / K2.6 join the model menu and are billed like the other cloud models.
  • FeatureDiscount coupons in the app: Enter a coupon code in the profile menu's Promotion field; the plan page shows the discount, a countdown and struck-through prices, and applies it at checkout or to a plan change.
  • ImprovementLiterature screening table: Paper, Abstract, TL;DR and Screening columns that adapt to the panel width.
  • ImprovementFigure colours: Choose a colour palette (Default, Warm, Cool, Pastel, Earth, Vivid, Navy) from the figure editor's style menu.
  • ImprovementFigure citations: Citation badges match the report's chips and open the same source card when clicked; older figures are updated when displayed.
  • ImprovementDocs address: Documentation lives at neutropic.io/docs.

v0.1.30

September 15, 2026

  • FeatureExcel preview: .xlsx / .xlsm / .xls uploads open in the right panel as a table with sheet tabs, like CSV.
  • FeatureInline figures in manuscripts: Figures are rendered in place with their captions where the text discusses them.
  • ImprovementReview card: After a repair pass the review opens with What was improved — verdict before and after, resolved, still-open and newly raised issues.
  • ImprovementLiterature screening: Each source shows its abstract and an extracted TL;DR next to the screening verdict.
  • ImprovementChat typography: Slightly taller line spacing for answers.
  • FixPlan card: Stays under the request it belongs to and no longer jumps under the next message.
  • FixFailed steps: A skipped tool step is shown as "skipped" with its explanation on expand, not as a red error.

v0.1.29

September 14, 2026

  • ImprovementNo clarifying surveys: Skills no longer ask questions before starting — they take the most plausible reading and state their assumptions. Only the approval card for hard-to-undo actions can pause a turn.
  • ImprovementExample projects in your language: The example projects shown match your UI language.
  • ImprovementPaper writing: Manuscripts follow IMRaD plus a Conclusion, with tables and figures placed inline.

v0.1.28

September 13, 2026

  • FeatureDocs site: GitBook-style documentation at /docs — section tree, on-page outline, ⌘K search, previous/next, 41 pages in English, Korean, Japanese and Chinese, plus a reference page for every skill.
  • FeatureChangelog pages: Release notes for the web app, agent & API and desktop app inside the docs — one page per product, every entry tagged Feature, Improvement or Fix, with a version outline.
  • FeatureInline docs assistant: Ask a question at the bottom of any docs page and get an answer right there, streamed and grounded in the documentation, with links to the pages it used and a "Continue in a chat" hand-off.
  • FeatureCopy page: Copy any docs page as Markdown or its link, or open it in ChatGPT/Claude.
  • ImprovementDocs layout: Slim top bar, light canvas with a white content card, section icons, system/light/dark theme toggle, and a mobile breadcrumb bar that opens the tree.
  • ImprovementAccount and settings screenshots: Sign-in, profile menu, language, promotion, Customize (General, Skills, Add skill, Connectors, Usage), delete account, plan, projects and the @ picker, captured in all four UI languages.
  • ImprovementMarkdown renderer: One renderer for blog and docs — anchored headings, ordered lists, callouts, tables, code blocks with copy, italics.
  • ImprovementTerms and Privacy Policy: Rewritten to cover plan and refund mechanics, acceptable use, AI-output responsibility, data inventory, processors and cross-border transfers, retention periods, your rights and remedies, and contacts — in English, Korean, Japanese and Chinese.
  • FeatureProduct introduction deck: A slide-by-slide product introduction at /intro-of-product — keyboard, click and swipe navigation, contents menu, full screen, shareable slide links, and print-to-PDF; in all four UI languages with matching screenshots.
  • FixLogin destination: Signing in from a deep link (for example /new?q=…) now returns to that page instead of the default new chat.

v0.1.27

September 12, 2026

  • FeatureLanding research workflow: The features section now walks through the six stages — literature, research design, data analysis, statistics, figures, writing — with real session captures for each.
  • FeatureUse-case walkthroughs: Eight blog posts (literature, hypothesis, experiment, data analysis, statistics, figures, writing, review) rewritten as step-by-step walkthroughs of one real session, with screenshots.
  • FeatureSkills catalog and guides: /skills rebuilt from the live catalog (59 skills); each workflow skill links to a guide post in the new Skills blog category.
  • FeatureLocalized blog and screenshots: All 31 posts in Korean, Japanese and Chinese; screenshots switch to captures taken in the matching UI language.
  • FeatureDesktop bridge: When a chat is attached to the desktop app, file and code tools run on the device; first-run folder setup, native file attach, @ autocomplete for local files, media pack and on-device storage controls.
  • FeatureMobile shell groundwork: Session persistence, safe-area layout, audio/video recording from the attach menu, push-token registration.
  • ImprovementSlash commands in your language: /research, /analyze and the other commands expand in the interface language, so reports and figures come out in that language.
  • FixArtifact file names: Artifacts created without an extension get the right one for their type, and the workspace renders extension-less Markdown correctly.
  • FixReview turns: A review that produced only structured data now also writes the readable review report.

v0.1.26

September 11, 2026

  • FeatureLive streaming: Answers stream token by token as the agent writes them.
  • FeatureAudio and video uploads: Attach .wav/.mp3/.m4a/.flac/.ogg/.aac/.aiff and .mp4/.mov/.webm/.avi/.mkv/.m4v for transcription, voice, facial expression, pose and rPPG analysis.
  • FeatureEEG analysis: Attach a multichannel EEG table for preprocessing, band power, topomaps, ERP, ERD/ERS and SSVEP.
  • FeatureGoogle Drive connector: Connect Drive under Customize → Connectors, control each tool (Always allow / Ask each time / Block), import files from the + menu, and approve tool use inside the chat. Hidden until Google app verification completes.
  • FeatureDiscount coupons: Discount codes can be entered at checkout.
  • FeatureStorage usage: Customize → Usage shows uploads + artifacts against the plan quota; plan cards list storage limits.
  • FeatureSkip the rest: Survey cards get a "Skip the rest — search now" button that fills unanswered questions with defaults.
  • ImprovementStatistics: Group comparisons choose parametric or non-parametric tests from assumption checks and report the rationale, effect sizes, post-hoc tests and descriptives.
  • ImprovementSlash menu: All top-level skills as aliases with descriptions; unknown /commands are caught before sending; commands that need a topic disable Send until one is typed.
  • ImprovementTurn banners: Step-limit stops with deliverables read as an informational notice, not a failure.
  • ImprovementList performance: Artifact cards load previews only in the viewport; artifacts, chats and projects paginate 50 at a time.
  • ImprovementSign-up consent: Separate required (terms, privacy) and optional (marketing) checkboxes, unchecked by default, with consent time and version recorded.
  • FixPanel auto-open: Helper artifacts (structured JSON, screening tables, descriptives) no longer open first when a turn ends.
  • FixStop and reconnect: Send is locked until the server confirms a stop; connection failures show a Retry button and keep your text.
  • FixHydration warning: Footer dot pattern moved to CSS, removing the React hydration mismatch on every page.

v0.1.25

September 10, 2026

  • ImprovementPaper writing: /write turns deliver the manuscript and its figures only; the reference list is built from the [n] citations in the text, so numbering and DOIs always match.
  • ImprovementEffort levels for writing: Lite, Basic and Deep produce progressively longer manuscripts with more figures; Basic and Deep add a review report.
  • ImprovementCitation badges everywhere: Manuscripts, design reports and review reports show [n] badges with reference cards and the source panel, like research reports.
  • FixExperiment design consistency: Sample sizes, effect sizes, α and test types in the power curve and expected-results figure are checked against the design report before saving.
  • FixEEG on numpy 2.4: Band power and PSD analyses work again.

v0.1.24

September 09, 2026

  • ImprovementArtifact language: Artifacts follow the language of the request more reliably.

v0.1.23

September 08, 2026

  • ImprovementIllustrations: Better scientific illustration quality.
  • ImprovementHero video: Updated landing hero video.

v0.1.22

September 07, 2026

  • ImprovementIllustrations: Illustration skill tuned.
  • FeatureSubscription alerts: Email alert when a subscription is upgraded.

v0.1.21

September 05, 2026

  • FeatureModel catalog 2026-09: Gemini 3.8 Flash (default), 3.1 Pro, 3.5 Flash-Lite; Claude Fable 5.1, Opus 5, Sonnet 5, Haiku 4.5; GPT-6 Astra and GPT-5.6 Sol/Terra/Luna.
  • FeatureDaily cap: Plan cards and the comparison table show the daily token cap and when it refills; the composer shows today's remaining tokens.
  • ImprovementLaTeX in chat: Better math rendering in chat responses.
  • FixToken figures: 1 credit now maps to ≈1,000 tokens (previous display was 4.5× too low).
  • FeatureRefund policy: Refund policy published.

v0.1.20

September 04, 2026

  • FeatureFigure editor: Select, drag, rename, add and connect nodes on server-rendered vector figures; group, align and edit properties; changes are saved as revisions.
  • FeatureVector figures: Figure artifacts render inline as SVG with SVG download and PNG export.
  • ImprovementPlan card placement: The agent plan appears under the latest request instead of at the bottom of the chat.
  • ImprovementToken display: Per-answer token count sits next to the copy button; a remaining-tokens gauge lives in the composer.
  • FixPicker scrolling: Keyboard navigation in the / and @ pickers keeps the selected item visible.

v0.1.19

September 03, 2026

  • FeatureFigure export: PDF and PPTX (editable objects) from the panel download menu, plus journal presets (single 90 mm, double 180 mm, poster, slide) and 300-dpi PNG.
  • FeatureVersion history: The artifact ⋮ menu lists revisions and restores earlier versions of text and figures.
  • FeatureEffort levels: Lite / Basic / Deep in the composer control search breadth, report length, review loops and step limits.
  • FeatureIllustrations in figures: Generate or pick an illustration for a figure node, with reference images from the chat.
  • ImprovementFigure editor: Resize handles, color inputs, alignment and distribution, snapping while dragging, full-page "Edit in workspace".
  • ImprovementCompressed tool results: Tool steps show a "compressed N%" badge when the model received a shortened result.

v0.1.18

September 01, 2026

  • FeatureProject knowledge panel: Evidence graph with search, synonym merging and a full-screen graph modal; artifacts list with filters.
  • FeatureStructured-output editor: Edit the fields of a <skill>_structured.json artifact and re-render the document.
  • FeaturePermissions: Customize → Permissions shows per-session approvals with one-click revoke.
  • ImprovementReview details: Review panels list issues with severity, location, problem and suggested fix; reproducibility shows source datasets and the agent run.
  • ImprovementExample project: Read-only for everyone, including the owner.
  • FixTool errors: Failed steps show the error code, whether it is retryable, and the recovery hint instead of disappearing.
  • FixArtifact loading: Failures show the reason with a Retry button instead of an endless spinner.

v0.1.17

August 31, 2026

  • Feature/study: Literature → design → analysis plan → integrated proposal in one turn, with stage progress.
  • FeatureSlash commands: /analyze, /stats, /hypothesis, /write, /review, /fact and /compact.
  • FeatureAgent plan panel: A collapsible step list with per-step status above the reasoning indicator.
  • ImprovementTurn restore: A notice explains when an interrupted turn was resumed from its checkpoint, or why it could not be.
  • ImprovementReview criteria: Review blocks show which criteria were applied.

v0.1.16

August 28, 2026

  • ImprovementFree tier: 500K managed tokens for Free accounts.

v0.1.15

August 27, 2026

  • FeatureProjects: Artifacts and chats linked at the project level; project deletion; an example project for new accounts.
  • FeatureSuggested tasks: Prompt suggestions on the new-chat screen.
  • ImprovementFollow-ups: Follow-up requests build on the previous answer.
  • FeatureMermaid diagrams: Diagram generation in artifacts.

v0.1.14

August 26, 2026

  • ImprovementResearch artifacts: Better artifact interface for research outputs.

v0.1.13

August 25, 2026

  • FeatureMobile menu: Hamburger navigation on phones.
  • ImprovementChat titles: Titles summarize the conversation instead of repeating the prompt.
  • ImprovementCSV and Markdown views: Better table and research-report rendering; research highlights restored.
  • ImprovementSettings language: The settings dialog follows the UI language.

v0.1.12

August 24, 2026

  • FeatureSettings → General: Profile and preferences.
  • ImprovementCredits display: Token and credit figures in chat, usage and plan views.
  • ImprovementOpen Graph images: Background image for link previews.

v0.1.11

August 23, 2026

  • FeaturePaper table: Literature searches produce a screening table ranked by citation score, journal score and similarity.
  • FeatureTransactional email: Sign-up and subscription confirmation emails.
  • ImprovementMobile views: Settings dialog, source modal and bottom sheets fit the phone screen; Google profile pictures shown.
  • FixInterrupted turns: A stopped turn is marked as such in the chat.
  • FixDesign reports: Reference annotations attach correctly; sample size matches the power curve.

v0.1.10

August 21, 2026

  • FeatureBlog and SEO: Blog section, metadata, sitemap and robots.
  • FeatureLanguages: Korean, Japanese and Chinese UI.
  • ImprovementDocument view: Markdown artifacts render as a typeset document with reading width, heading hierarchy, tables and quotes.
  • ImprovementCitation cards: [n] badges with hover reference cards in research reports.
  • ImprovementArtifact names: Files show their real file names, including extensions.

v0.1.9

August 20, 2026

  • FeatureDesktop app shell: App mode hides marketing chrome; Google sign-in through the system browser with a deep-link callback.
  • ImprovementGoogle account switching: Sign-in always shows the account chooser.
  • FixGhost sessions: A revoked server session no longer leaves the UI signed in with failing requests.
  • FixUsage gauge: The plan gauge reads used/granted instead of the remaining balance.

v0.1.8

August 19, 2026

  • FeatureBilling: Checkout connected; pricing and token policies revised.
  • ImprovementLogo: Updated brand mark.

v0.1.7

August 18, 2026

  • ImprovementResearch skills: Search and report quality.

v0.1.6

August 14, 2026

  • FeatureGemini 3.x: Gemini 3.6 Flash (default), 3.1 Pro and 3.5 Flash-Lite.
  • ImprovementReport references: Source composition and per-database contribution shown as badges in the References section.
  • FixEmpty turns: Tool calls written as text and questions written after the clarifying budget no longer end a turn without a result.
  • FixGemini function calls: Restored the thought signature Gemini 3.x requires on tool calls.

v0.1.5

August 12, 2026

  • ImprovementResearch skills: Search and report quality.

v0.1.4

August 11, 2026

  • ImprovementResearch skills: Search and report quality.
  • ImprovementLogo: Updated brand mark.

v0.1.3

August 09, 2026

  • ImprovementLogo: Updated brand mark.

v0.1.2

August 06, 2026

  • FeatureResearch tracks: Papers, patents and web as separate tracks composed by /research; /patents searches Google Patents, PatentsView and KIPRIS.
  • ImprovementSkill keys: Hyphenated skill keys (experiment-design) with legacy aliases.

v0.1.1

August 05, 2026

  • ImprovementResearch skill: First round of research-skill improvements.

v0.1.0

July 28, 2026

  • FeatureNeutropic 2.0: Initial draft of the new web app.