Portail

Version log

Every Portail release, written down.

Each release gets a public changelog post, product screenshots, structured data, and a sitemap entry so the record is easy to find later.

v2.13.02026-11-142 screenshots

Portail calls: meet right inside Portail, with Portail in the call

Meets now has a third way in. Until today, Portail could listen in the room through your microphone, and it was getting ready to join Google Meet, Zoom and Teams calls as a participant. Now you don't need another call app at all. Open Meets, press Start a call, and send the link. Up to six people talk on video, share their screens and type in a chat everyone sees, and Portail sits in the call as its own tile, taking notes the whole time. Because every person's own browser transcribes their own microphone, every line of the transcript carries the real name of whoever said it. There's no guessing who spoke and no echo from speakers, and when you mute yourself, Portail stops hearing you too. Anyone in the call can say "Portail," and a question, or type it in the chat, and Portail answers in the chat for everyone, using only what was said in the meeting. Ask it for a plan, a checklist or a follow-up email and it drafts one on the host's meeting page. The notes fill in as you go: a summary, the decisions, and action items with owners. Everyone in the call can read the transcript and notes, and the host's private Ask Portail chat and docs stay private. Joining takes a signed-in Portail account, and everyone sees a short notice before they join: Portail transcribes the call, keeps the text and notes, and never keeps the audio. The calls run on Portail's own call server, not a third party's, and nothing is ever recorded. The host stays in charge: copy the invite link, make a new link so the old one stops working, remove someone from the call, or end it for everyone. Every plan can host. Calls on the free plan run for up to an hour, and paid plans get up to eight hours. The host's credits pay for transcription and answers, so guests never spend their own. When the call ends, the host gets the usual meeting page with the full transcript and notes, plus the recap email if it's turned on.

1047 words · minor release

Read post
v2.12.12026-11-14

Meets hears your Zoom and Teams calls, no bot required

Portail Meets could already listen in the room and pick up a call running in another browser tab. Now it can hear a call in any app on your computer. Turn on "Also hear the call", share your entire screen with "Share system audio" on, and Portail hears the Zoom or Teams app the same way you do, without joining the call as a participant. It works in Chrome and Edge on Windows, and on macOS 14.2 or later. Sharing just the call's tab still works for calls in the browser. The transcript now tells the two sides apart: what your microphone hears is labelled with your name, and what comes from the call is labelled Others, so the notes and action items know who said what. If you're not wearing headphones, your microphone hears the call through your speakers too. Portail now spots that and drops the repeated line, first in the browser and again on the server, so the transcript doesn't show every sentence twice. Headphones still give the cleanest transcript. With no bot in the call, nothing posts a recording notice there for you, so the meeting page has a new Copy notice for the call button: one short message you paste into the call's chat so everyone knows Portail is taking notes and that it keeps the transcript, never the audio. In-person meetings with just the microphone work exactly as before, and nothing changes in price: transcription is billed the same way it always was, on every plan.

673 words · patch release

Read post
v2.12.02026-11-134 screenshots

Meets: Portail sits in your meetings, takes the notes, and answers when you say its name

Portail now has a Meets page. Press Start listening and Portail keeps a live transcript of the meeting, writes the notes as people talk (a short summary, the decisions, and the action items with who owns them), and answers when someone says "Portail," followed by a question. It works for a meeting around a table and for any call app, because it listens through your browser; in Chrome or Edge it can also hear a call running in another tab. Ask out loud or from the meeting page for a doc, a plan, a timeline, a checklist, a follow-up email or the next agenda, and it shows up on the page before the meeting is over. When the meeting ends, the final notes land in your inbox. Later, ask about any past meeting from an ordinary chat: what did we decide in Tuesday's sync? Portail keeps the transcript, the notes and what it made, never the audio, and you accept a short recording notice before the first meeting, because everyone in the room should know. Sending Portail into a Google Meet, Zoom or Teams call as its own participant, from a link or straight from your calendar, is coming soon on paid plans and is marked Soon on the page. Meets is in beta. Tell us what it hears wrong.

1061 words · minor release

Read post
v2.11.42026-11-12

Google Calendar in chat: move and cancel single events, no surprise Meet links

The Google Calendar connector got a round of fixes. Events Portail added to your calendar no longer come with a Google Meet link you never asked for, and the chat now gets the event's link back, so it can show you where the event landed. You can ask it to move, rename or cancel one specific event, and it changes only that one. Before, the only way to remove something was to clear a whole stretch of time, and that also caught events that merely overlapped it, like an all-day event or a meeting running into the hour. Clearing a range now removes only what fits inside it, and does it faster. When a busy week has more events than one answer can show, the chat says so instead of presenting the first 20 as everything. Replies in chat also start a little sooner when you're signed in, because Portail no longer checks your connected apps from scratch on every message.

596 words · patch release

Read post
v2.11.32026-11-111 screenshots

Chinese, Japanese and Arabic PDFs, plus file previews

Designed PDFs now handle the languages that used to come out as rows of empty boxes. Chinese, Japanese, Arabic and Hebrew text gets a matching Noto font letter by letter, so a report that mixes English product names with Chinese paragraphs, or Arabic sentences with numbers, reads the way it should. Arabic and Hebrew paragraphs run right to left, with bullets on the right, table columns in reading order and brackets facing the right way. Chinese and Japanese lines wrap between characters instead of spilling off the page, and a Japanese file uses Japanese character shapes rather than Chinese ones. The download card in chat also shows a small picture of the first slide or page of the decks, PDFs and Word files Portail designs for you, so you can see what you got before you open it. And when a deck is so long that it hits the 80-slide limit, the card says how many slides were left out, and the model knows it too, so it can offer to split the content into shorter decks.

681 words · patch release

Read post
v2.11.22026-11-10

New chats stay in your chat list

A chat you had just started, often one opened from the Finance page, could disappear from the chat list a minute later. Its save had failed quietly, for example when the page had made a lot of requests in a short time, and the list then reloaded from the server without it. Now a chat stays in the list while its save is still on its way, a failed save is tried again on every refresh, and saving chats no longer competes with the rest of the app for the same request budget.

459 words · patch release

Read post
v2.11.12026-11-10

Finance now reads PDF bank statements

Personal Finance used to take statements only as CSV or Excel files. Plenty of banks hand you a PDF and nothing else, or bury the spreadsheet export three menus deep. You can now import the PDF straight from your bank's app or website, and Portail rebuilds the table of transactions from it: dates, descriptions, amounts, and the running balance when the statement has one. It works with English and Russian statements, with one amount column or separate money in and money out columns, with descriptions that wrap onto a second line, and with tables that restart on every page. Sberbank-style statements, where each operation takes two lines and only income carries a plus sign, come through with every purchase counted once and the salary counted as income. Nothing about this uses an AI model: the file is read the same careful, rule-based way spreadsheets already were, and it stays private to your account. Scanned statements and password-protected PDFs get a clear message saying what to do instead, since there is no text in a scan to read and a locked file can't be opened. A PDF that holds no transaction table at all, such as a letter from your bank, is turned away with a pointer to the CSV or Excel export. The 4 MB limit per file still applies, so for a long history, download a few shorter periods and import them one after another. Overlapping statements are fine: transactions already imported are recognized and skipped, so nothing is counted twice. Imports keep working on every plan, free included, and you can keep up to ten statements at a time.

665 words · patch release

Read post
v2.11.02026-11-103 screenshots

Finance: your accounts, spending and bills in one place, and a chat that knows the numbers

Portail now has a Finance page. Import a CSV or Excel statement from any bank in any country, and Portail adds up where your money goes, finds the subscriptions and bills that repeat, lists what's due before your next paycheck, and points out anything worth a second look: a charge that went through twice, a fee, a subscription that quietly got more expensive, a month that ran well over the usual. Then ask about it in plain words. Where did my money go last month? Which subscriptions am I paying for? Am I spending more than I earn? The answers come from your real transactions, not guesses, in English or Russian, with any model you like. It is read-only from end to end: Portail can see balances and transactions, but nothing here can move, pay, transfer or invest money. You can remove a statement whenever you want, which deletes what Portail kept from it. Chats that read your accounts are never saved to memory. Importing statements works on every plan, Free included. Linking your bank directly, through Plaid, is coming soon and is marked Soon on the page. Finance is in beta, so tell us what it gets wrong.

1058 words · minor release

Read post
v2.10.12026-11-09

Hold to dictate now tells you when to let go

When there's already text in the box, holding the send button switches it to dictation, and you start talking once you lift your finger. Until now the only sign that the switch had happened was the icon quietly changing from an arrow to a sound wave, which is easy to miss when your thumb is covering the button. So people either let go too early and sent a half-written message, or held on for ages wondering if anything was going to happen. Now the moment the hold arms, the button pops up a little bigger, the sound wave spins into place, and a soft ring keeps rippling out from it for as long as you keep holding. That ring is the cue: let go now and the mic opens. On phones, letting go also gives you a short tap of haptic feedback, so you can feel the switch without looking down. Android uses the standard vibration API. On iPhone, Safari has no vibration API, so we borrow the tick iOS plays when you flip a switch, which works on iOS 17.4 and later. If you slide your finger off the button before letting go, nothing happens, no message and no buzz, same as before. On a laptop with a mouse you get the animation but no vibration, since there's nothing to vibrate. People who turned on reduced motion in their system settings get the change of icon without the bounce or the ripple. Tapping send normally works exactly the way it did: a quick press still sends right away, and the hold still takes a bit under half a second to arm.

678 words · patch release

Read post
v2.10.02026-11-061 screenshots

Bring your own OpenRouter key

A new way to keep chatting without waiting for a reset: your own key. Open Settings, choose Your own key, paste an OpenRouter API key, and your chats are paid for there instead of with Portail credits. Chat, Council and Ratchet all use it, and because you are the one paying, models above your plan's line open up too. We check the key with OpenRouter before we save it, encrypt it before it is stored, and only ever send it to OpenRouter. After you save it you never see it again, just its last four characters, and you can replace it, switch it off or delete it whenever you like. Set a spending limit on the key at OpenRouter and it can never cost more than you mean it to. If the key is refused or runs out of credit, you decide what happens next. By default that one message is paid with Portail credits and is marked so you can tell, and a key OpenRouter has refused is switched off so it stops getting in the way. Turn the fallback off and you simply see a plain message saying what is wrong with the key. Images, Canvas, Telegram and scheduled tasks keep using Portail credits. Everything is in English and Russian.

836 words · minor release

Read post
v2.9.22026-11-06

Watchers keep their rhythm

A few fixes to how watchers pace their checks and read their own answers. Pausing and resuming a watcher used to start a fresh check the moment it came back, and so did saving it again with a new check speed. Now a watcher that comes back waits until its usual interval has passed since its last check, so a watcher set to check every 3 hours still checks every 3 hours, however often you pause it. If it has been asleep longer than that, it checks right away, as before. Changing what it watches for still counts as a new watch and checks right away too, because a condition that is already true is news. The next check time shown after you edit, pause or resume a watcher is now the one that will really happen. Before, a watcher slowed down by a plan change could show a time it was going to skip, both in settings and in the Telegram /tasks list. The time it saves after each check had the same problem and is fixed too. Last, a watcher that found something and put its whole headline on the same line as its alert, with a long headline, was read as giving no answer at all. That counted as a failed check, and three in a row paused the watcher. Those alerts now come through with their headline as written, and the watcher keeps running.

640 words · patch release

Read post
v2.9.12026-11-06

Start a watcher right from the chat

Watchers no longer have to be set up in settings. Ask Portail something like "tell me when the XM5 drops under $280" or "let me know when my university posts exam dates", and it answers what it can right now, then offers to keep an eye on it. A Start watching card appears under the reply with the exact condition it would watch for, how often it will check on your plan and where it will tell you. One tap starts it. Customize opens the full watcher form with the condition filled in, if you want a different speed, delivery or model. Nothing is set up until you tap: Portail only proposes, and the watcher goes through the same plan and limit checks as one you create in settings. New watchers start on a fast, cheap model, because every check uses a few credits. If you've reached your plan's limit on watchers and tasks, the card says so and points you to Manage watchers or a bigger plan instead. Every message you send also has a new Watch for this button next to Schedule this, which opens a watcher with your message as the condition. Offers never appear in incognito chats, and the card remembers that you started a watcher, so it shows as done after a reload.

619 words · patch release

Read post
v2.9.02026-11-062 screenshots

Watchers: Portail keeps an eye on things and only speaks up when they change

Tell Portail what you're waiting for, in plain words, and it watches for you: a price dropping under your number, an email from your university about exam dates, a company finally announcing its event, real news on a topic you follow. Portail checks the web and your connected apps on its own and stays completely quiet until it finds something new that matches. Then it sends you one short message on Telegram or by email with what changed, the numbers that matter and the link. That silence is the point. A watcher that pings you every hour to say nothing happened is a newsletter, and newsletters get muted. Free checks once a day, Pro every 3 hours, Plus and Ultra every hour. Find them in Settings > Watchers & tasks.

1079 words · minor release

Read post
v2.8.132026-11-05

Fresh models: GPT-6.1 Terra, Kimi K3.1 Pro, Gemini Omni Flash

We pulled the latest model list from our provider. Thirteen chat models are new in the picker, among them GPT-6.1 Terra and Terra Pro, Kimi K3.1 Pro, Gemini Omni 1.1 Flash and 1.2 Flash, Tencent's Hy4 and Z.ai's GLM 5.3 FlashX Legal, plus a handful of small, cheap models for quick jobs. Canvas gets seven new image models: Krea 2 in Large, Medium and Medium Turbo, Krea's K3 Edit, GPT Image 2.6 Flare and Sunburst, and Riverflow V2.6 Pro. None of them are switched on by default in Canvas, so tick the ones you want to compare. Twenty-six chat models are no longer offered by our provider, including OpenAI's o1, o3-mini, o4-mini, GPT-4 Turbo and GPT-3.5, the older MiniMax and Hermes 3 models, the Poolside Laguna family, and a few free versions of models. If a chat of yours was using one of them, it keeps working: your next message goes to the closest model that is still around, from the same maker where one exists. o4-mini and o3-mini move to GPT-5 mini, o1 moves to o3, GPT-4 Turbo moves to GPT-4o, and the Laguna coding models move to Devstral. Chats on a free model move to another free model, so they stay free. Prices for every model still come straight from what the provider charges us, the same as before, and your plan is unchanged. A chat that moves to a new model is billed at that model's rate from your next message on. One Canvas model, Grok Imagine Image Quality, is gone as well. Grok Imagine Image 2.0 is still there and still on by default.

646 words · patch release

Read post
v2.8.122026-11-04

Shorter sign-up, and photos one tap closer

Signing up used to mean answering eight or nine questions before your first message: what you do, what you're here for, which other AIs you've used, where you heard about us, a starter prompt, and an offer to bring your memory over from ChatGPT or Claude. That's fine if you live in AI tools all day, but plenty of people just want to get in and ask something. So sign-up now asks for your name, your age and your account details, then one question: how familiar are you with AI assistants? If you pick "I use them every day", you get the full set of questions, because that's where they pay off: the chat learns your role and goals, and you can import what your old assistant knew about you. Pick anything else and you go straight to the chat. Nothing is lost by skipping, since Portail still learns about you as you talk to it, and Enhanced Context memory works the same either way. The second change is for phones. Attaching a photo on Android used to take four steps: the plus button, Files, the file manager, then a hunt for the gallery. The plus menu now has a Photos item above Files. On Android it opens the photo picker directly, and on iPhone it opens your photo library first. Files still takes every type we support, from PDFs and spreadsheets to source code, and pasting an image into the box works as before.

636 words · patch release

Read post
v2.8.112026-11-02

Reminder emails that know when to stop, now in Russian

If you sign up and never ask anything, or you go quiet for a while, Portail now sends a short note to bring you back to where you were. There are at most three of them, spread over a few weeks, and the last one says it is the last one. You never get one on the same day as a Portail tip email, they skip anyone who unsubscribed, and nobody hears from us after 45 quiet days. If you left a conversation open, the note names it, so you can pick it up in one click with any model you like. These notes now come in Russian for people who use Portail in Russian. We go by the language you write in, then the language your Portail screen is set to, then the language the Telegram bot speaks to you in, and fall back to English when none of those says otherwise. The footer, the unsubscribe link and the page you land on after unsubscribing are translated too, so a Russian note doesn't arrive wrapped in English. We also fixed a line about credits that was only half right. Free credits still reset each month, but they also refill every few hours and every week, and the note now says exactly that instead of promising that whatever you spent is already back. One more fix matters more than all of the above. Signing in to Portail could quietly undo an unsubscribe, so someone who had turned off tip emails could start getting them again after their next sign-in. The same slip sometimes sent a fresh welcome email instead of a sign-in notice. Both are fixed: an unsubscribe now stays put no matter how often you sign in, and the welcome email only ever arrives once.

715 words · patch release

Read post
v2.8.102026-11-01

Portail in Russian, with its own web addresses

The landing page, the model pages, the comparisons and the feature pages now exist in Russian at their own addresses, such as portail.cc/ru and portail.cc/ru/compare. Before this, the Russian text only appeared after the page loaded, so Russian-language search could not find any of it. Now the page arrives in Russian from the first byte, with a Russian title, description and FAQ, and search engines are told which version belongs to which language. The EN and RU switch on these pages now moves between the two addresses, so what you see, the link you copy and the page you share always match. Choosing a Russian page also remembers Russian for the rest of the site. Nothing changes for the chat itself, and English addresses work exactly as before.

501 words · patch release

Read post
v2.8.92026-11-01

A sharper headline font on the landing page

Landing page headings now use Geologica instead of the same Inter as everything else, in English and Russian. Body text, the chat and the rest of the product are unchanged.

378 words · patch release

Read post
v2.8.82026-11-01

The landing composer shows the lineup

While the landing page's ask box sits idle, GPT-5.5, Claude Sonnet 5 and Gemini 3.5 Flash take turns lighting up beside it, a quiet hint that one thread works with any of them. It stops as soon as you click, type or pick a model, and stays off if you prefer reduced motion.

406 words · patch release

Read post
v2.8.72026-11-01

A calmer, more readable landing page

The landing page lost its floating logo field and scrolling model ticker. A plain row of model names replaces them, the small grey labels are now dark enough to read, and the feature grid gives Connectors the space it deserves.

408 words · patch release

Read post
v2.8.62026-10-31

Link Ratchet while you sign up

Linking Ratchet to a brand-new Portail account used to be a race. Running ratchet --link opens Portail in your browser, and if you didn't have an account yet, you signed up first. Sign-up asks a handful of questions before it lets you in: your name, your age, your account and the code we email you, and then a few more about your role, your goals and how you use AI. Ratchet only waits a few minutes for the browser to come back, so anyone who took their time on those questions finished sign-up to find that Ratchet had stopped listening, and the link went nowhere. Now, when you sign up on the way to linking Ratchet, Portail sends you back to finish the link as soon as your account exists and your email is confirmed. The rest of the questions wait for you, and the chat asks them the next time you open it. The same goes for signing up to Portail Agents, and for apps that ask you to approve access to your Portail account: you see the approval screen as soon as your account is ready instead of after the whole questionnaire. Nothing about age checks changed. Sign-up still asks your age before it creates the account, teens still get the AI street smarts guide first, and the age bracket is saved before you leave for Ratchet, so the teen guidance applies there too. The next Ratchet release also waits much longer for the browser to come back, up to half an hour, and you can still cancel at any time with ctrl+c or the Cancel button in the desktop app. If you already have a Portail account, nothing changes: you sign in and approve the link as before.

719 words · patch release

Read post
v2.8.52026-10-31

Smoother log in and sign up

A round of fixes for logging in and creating an account. The biggest one: if you came to Portail from Ratchet, an app asking to connect, or a gift card, and made a new account along the way, you ended up in the chat instead of back where you started. Now you land where you were going. Sign-up errors also read properly in Russian, a code screen that has run out of time sends you back to the form with your email still filled in, and the Back button no longer bounces you off the log in page.

621 words · patch release

Read post
v2.8.42026-10-31

Russian requests no longer turn into pictures by mistake

Portail decides on its own when a message is asking for a picture, so you don't have to switch modes: on a model that can draw, a request like "create an image of a fox" goes straight to image generation. In the last release that check learned Russian, but it matched too loosely. On an image-capable model, everyday requests such as "сделай план на март" or "создай карту сайта" could be read as picture requests, because a few letters inside ordinary words looked like the start of a word for "art" or "picture". That sent a text question to image generation and charged for an image nobody asked for. This update fixes the check so it only recognizes picture words at the start of a word, and it now also understands "нарисуй" (draw) on image models, which it missed before. Carousels got two small fixes too: a page whose picture comes from an outside link is left out instead of showing only its title, and the page number and swipe arrow stay visible on bright custom backgrounds.

617 words · patch release

Read post
v2.8.32026-10-31

Student discounts find the right school more often

Three small fixes to how portail.cc/students matches the school name a student types. First, Uzbek names written with the acute accent as an apostrophe, like o´rta or Ulug´bek, now read the same as o'rta and Ulugbek. Before, that one character split the word in two and the school wasn't found at all. Second, a bare name that fits a school and a high school with the same name now asks which one you mean. If Sodiq School and Sodiq High School both offer codes, typing just Sodiq used to quietly hand out a Sodiq School code. Now the page asks for the full name, the same way it already did when the two places were a school and an academy. Typing the full name, Sodiq School or Sodiq High, still goes straight to the right one. Third, the other names a school accepts are now all checked. If a school lists a short name such as Sodiq as one of its accepted names, typing exactly that short name now finds that school, even when another place called Sodiq Academy also offers codes. Before, only the school's main name was compared, so the student was asked for the full name even though they had typed one the school accepts. If two different places both list the same short name, the page still asks which one you mean rather than picking one. None of this changes who can get a code: each school's codes still only go to students who name that school, and a different number in the name never matches.

650 words · patch release

Read post
v2.8.22026-10-29

Student discounts, through your school

Schools that partner with Portail can now hand their students a discount without passing around a shared code. A student opens portail.cc/students, signs in, types their name and their school, and gets a personal single-use code for the plans their school arranged. The school name doesn't have to be typed perfectly: a missing space, a dropped letter or a swapped pair still finds the right school, while a different number (School 5 versus School 6) never does. Each account gets one code, and coming back to the page shows the same code again instead of a new one. From the code card a student can go straight to checkout with the discount already applied, or copy the code and type it into the checkout page. Codes only work on the plans and billing periods the school chose, and if a student tries a plan the code doesn't cover, Portail says so before checkout opens, so nobody ends up paying full price by accident. Behind the scenes every code is its own one-time discount, so a code that gets shared in a group chat works exactly once.

615 words · patch release

Read post
v2.8.12026-10-26

Portail fits your phone better

A pass over the small things that make a web app feel off on a phone. The chat now sizes itself to the visible screen, so the message box no longer hides behind the browser's address bar. Copy, edit, rename and delete used to appear only on mouse hover, which left touchscreens out; they are now always visible there. Sidebar rows and header buttons are taller, tall pop-ups scroll inside the screen instead of being cut off, form fields no longer make iPhones zoom the page, and wide tables in answers scroll sideways instead of squeezing every column.

493 words · patch release

Read post
v2.8.02026-10-245 screenshots

Slides, PDFs and carousels that look designed

Ask any model in Portail for a presentation, a PDF report, a resume or a Word document, and the file now comes back properly designed instead of as plain text on a white page. Portail picks a theme, lays out every slide or page, sizes the text so nothing spills over, and turns your numbers into real charts, big-number cards and clean tables. This works the same with every model, including the free and budget ones, because the design is done by Portail, not by the model: the model writes the content, Portail makes it look good. A one-line request like "make a presentation about coffee" gets a cover, varied layouts and a closing slide. You can still steer everything: name a theme, give it your brand color, ask for a gradient background or shapes on a slide. When you want changes, just say so in the chat and you get an updated file with the design intact. There's a new format too: Instagram and LinkedIn carousels, as swipeable 4:5 pages. And on phones, a row of shortcuts above the chat box gets you to slides, reports, websites, images, carousels, voice, scheduled tasks and connectors in one tap.

1122 words · minor release

Read post
v2.7.92026-10-24

Make the theme yours

Settings > Customization now has a custom theme builder. Pick a preset (Midnight, Forest, Ember, Plum, Sand or Mint), or choose your own background and accent color and watch the whole app follow along. Surfaces, borders and text are worked out from your two colors so everything stays readable, Like what you made? Save it as a preset of your own. The choice sticks on that device until you reset it.

503 words · patch release

Read post
v2.7.82026-10-24

A clearer Share button in the chat header

The chat header used to show Upgrade, a lone share icon behind a thin divider, and Incognito. Two of the three had a name and one did not, so Share looked like an afterthought. Share now carries its own label and matches Incognito in style, and the stray divider is gone, so the three controls read as one tidy row. On small screens the labels collapse to icons so the header stays compact.

441 words · patch release

Read post
v2.7.72026-10-24

Free plan: pricey models now fall back instead of failing

If you are on the free plan and pick a model that needs a paid plan, Portail no longer stops you with an error. It answers with a lighter model from the same maker instead and marks the reply as switched, so you can see what happened and upgrade if you want the original.

422 words · patch release

Read post
v2.7.62026-10-24

Finances lists every paid account

The admin Finances tab now names every account on a paid plan that has no live subscription, so customers sold by hand no longer go unseen.

373 words · patch release

Read post
v2.7.52026-10-24

The priciest models are now Ultra-only

Plus still unlocks the flagship models, but now stops at models priced above $25 per million tokens (input and output combined). Those, such as Claude Opus 5 and GPT-5.5, are available on Ultra. Plus members keep every other flagship, including Claude Opus 5.5, and the picker shows which models need Ultra.

417 words · patch release

Read post
v2.7.42026-10-23

Pro no longer has a free trial

Pro now starts billing the moment you subscribe. The 7-day free trial on monthly Pro has been removed from checkout, the pricing page, the upgrade window and the refund policy. The Free plan is unchanged and stays free with no card needed, and anyone already in a trial keeps it until it ends.

412 words · patch release

Read post
v2.7.32026-10-23

Auto now chooses from the whole model catalogue

Auto mode used to pick from a short, hand-kept list of about fifteen models, so most of the catalogue never got a turn and new models had to be added by hand before Auto would use them. It now considers every chat model Portail offers. For each message it still decides how demanding the request is: small talk, an everyday question, a real question that has to be right, code, or hard multi-step work. Then it scores every model that fits. The score weighs the model's size class, whether it is the newest release in its own family, how well its price suits the job, whether it reasons, and who makes it. Quick questions go to small, fast models, code goes to code-capable ones, and hard problems go to the strongest models your plan includes. Older generations lose to their successors, so a newly released model starts winning the moment it appears in the catalogue. Auto still respects your plan, so Free and Pro are never routed to a flagship model. Images only go to models that can see them, and long conversations only go to models with enough room to hold them. The most expensive reasoning models stay manual-only, because Auto spends your credits without asking. You can still pick any model yourself from the picker, and the Auto badge still shows which one answered.

635 words · patch release

Read post
v2.7.22026-10-23

Flagship models now start at Plus

The most expensive models, such as Claude Opus and GPT-5.5, are now part of Plus and Ultra. Free and Pro still reach every other model in the picker, including Sonnet, Gemini Pro, Grok and the rest. Flagship models show a small lock in the model picker on plans that can't use them, and tapping one opens the upgrade options. If a conversation was still set to a flagship model, Portail switches it back to the default and tells you why, instead of failing quietly. The same rule applies everywhere you chat: the web app, Telegram, scheduled tasks and Auto mode, which now picks only from models your plan can use. Ratchet follows the same rule when it runs on Portail credits, so a linked Free or Pro account needs to pick a non-flagship model there. We also refreshed the model catalogue: it adds dozens of new models and retires 25 older ones, and chats that were set to a retired model move to its closest successor.

572 words · patch release

Read post
v2.7.12026-10-17

Ratchet's desktop app, one click from code.portail.cc

The Ratchet site now has a Desktop app card in its Install section. The Linux build downloads straight from the page, checked against a SHA-256 before it is served, so a swapped or damaged file is refused instead of handed to you. It needs no Python, adds Ratchet to your app menu, and shares keys, sessions and your Portail link with the terminal app, so a conversation started in one picks up in the other. The Windows build is still being packaged, so its button reads not available yet instead of linking somewhere that does not work. The terminal install steps are still right below it, and the FAQ and the page's search metadata now mention the desktop app too.

498 words · patch release

Read post
v2.7.02026-10-151 screenshots

Talk, and it types

Voice mode is gone. The same sound-wave button in the chat box now does dictation: tap it, talk, and your words show up in the box while you speak, with steady background noise kept out. Nothing sends until you say so.

825 words · minor release

Read post
v2.6.282026-10-15

A page for every model, matchup and feature

Portail now has public pages for the questions people actually type into a search box: which AI is better, where to use a given model without its own subscription, and what one chat can do with your PDFs, inbox and calendar. Every model family gets a page with its full lineup read straight from the live catalog, so the numbers on it are never older than the catalog itself. Comparisons like ChatGPT vs Claude stay honest. They describe what each family tends to be good at, show the live counts side by side, and then point at the only test that settles it: your own prompt, asked of both, in one thread. Everything is in English and Russian, and it all lives one click from the footer.

638 words · patch release

Read post
v2.6.272026-10-15

Research that finishes the job

Ask a model to dig deeper and sometimes it would run one search, write "Next I'm checking independent sources", and then stop. The reply ended there, with a promise and no follow-up, and you had to push it again yourself. The model had described its next search instead of running it, and nothing told it that once it stops writing, nothing else happens. Two changes close that gap. First, every model with web search on is now told plainly that the reply ends when it stops writing. So when a question needs several searches, it runs them one after another in the same reply, and when the first results are weak (thin, off topic, or mostly vendor marketing) it searches again with a sharper query instead of narrating what it would do. After each search it also knows how many searches it has left. Second, Portail now notices when a reply that has already searched ends by announcing more research, in English or Russian, and sends it back so the model either runs that search or wraps up properly. You see the reply carry on under its last line. Questions and offers like "Want me to look into pricing next?" are left alone, since those are a fine way to end, and replies that never searched are never sent back. Each reply can also run up to five searches now instead of three, which gives research questions room to check more than one angle. Credits are charged the same way as before, and a reply is sent back at most twice, so it can never loop. The same fix applies in Telegram and in scheduled tasks.

648 words · patch release

Read post
v2.6.262026-10-14

Models that actually read your files

Attaching a file to a chat works much better now, whichever model you pick. Long files used to stop after the first stretch of text, and the model had no way to see the rest, so it would answer about a 200-page PDF having read only the first few pages. Now the model sees the beginning, knows exactly which pages or lines it hasn't seen yet, and reads them on its own when your question needs them: it can jump to a page, open one sheet of a workbook, or search a long log for the line that matters. You can watch it happen: each read shows up above the reply, like "Reading report.pdf, pages 12–14". Spreadsheets arrive with their sheet names, column letters, row numbers and formulas, so you can ask what is in C7 or why the total in B12 is wrong and get a straight answer. PDFs are numbered by page. Scanned PDFs, which used to arrive empty, are now shown to image-capable models as pictures of the pages. Files Portail made for you earlier in a chat are no longer forgotten: ask it to change the script or the spreadsheet it just gave you and it edits the real file instead of rewriting it from memory. Spreadsheets it creates now hold real numbers and working formulas, and can have more than one sheet. You can also attach code and config files directly: Python, JavaScript, TypeScript, HTML, CSS, Java, Go, Rust, SQL, YAML, logs and more. And if you send an image to a model that can't see images, it now tells you to switch models instead of answering as if nothing was attached.

687 words · patch release

Read post
v2.6.252026-10-13

Edit your PDFs in chat

Attach a PDF to a chat and ask for changes in plain words. Portail can now replace text, delete, rotate and reorder pages, pull out a few pages into a new file, merge in another PDF, stamp a watermark, add page numbers, add a line of text anywhere on a page, and fill in form fields. You get the edited copy as a normal download card, and your original upload is never touched. Because the copy is a new file, you can keep going in the same chat: ask for another change and Portail edits the version it just made. Edits are checked step by step. If a piece of text can't be found, for example because the page is a scan, Portail tells you which step failed instead of pretending it worked. One honest caveat: replaced text is drawn over the old text in a standard font, so fonts and colors can differ from the original, and covering text this way is not a secure redaction, since the hidden text can still be selected in the file. Portail says so whenever you ask it to hide something sensitive.

577 words · patch release

Read post
v2.6.242026-10-12

Watch the model work, step by step

When a model thinks or uses a tool before answering, you now see it happen in the order it happens. Before, all the thinking sat in one folded box and every tool call was grouped by type above the reply, so you could not tell what came first. Now each step appears the moment it starts. The model's thinking opens by itself and streams its text as it is written. When the model moves on, for example to search the web, look something up in your memory, or read your calendar, that thinking folds shut into a short "Thought for 3s" style line, and the next step shows up right below it with its own animated status. If the model thinks again after a tool call, that becomes a new step in the right place instead of being glued to the first one. You can still click any thinking line to open or close it whenever you like. Older conversations keep the layout they were saved with, so nothing you wrote before changes or disappears. Nothing about credits, plans or privacy changes: the same thinking and tool details that were shown before are shown now, just in a clearer order and in real time.

574 words · patch release

Read post
v2.6.232026-10-12

Fourteen more apps in Connectors

Portail's chat can now work with fourteen more apps from Settings > Connectors: Trello, Linear, Asana, Todoist, Google Drive, Outlook, Dropbox, HubSpot, Jira, Airtable, Calendly, Zoom, LinkedIn, and ClickUp. Link an account once through the same managed sign-in the other connectors use, then ask in plain words. Add a card to a Trello list, file a Linear issue, find a PDF in Drive, check what is on your Outlook calendar tomorrow, schedule a Zoom call, or move a Jira ticket to Done. Reads run straight away, and anything that sends, posts, or changes something on your account goes through the same flow as Gmail and Slack today. Scheduled tasks can use the new apps too, and their write actions stay behind the same per-task opt-in. The list is sorted so the apps you have linked come first, then the ones you can link now. Portail still never stores your passwords or access tokens; the sign-in lives with our connection provider, and you can disconnect any app from the same screen at any time.

569 words · patch release

Read post
v2.6.222026-10-12

Pickers that look like Portail

Every dropdown, time field and date field in scheduled tasks and the admin panel is now a Portail-styled picker instead of the one your browser draws. Pick a weekday, a model, a time or a one-off date from a menu that matches the rest of the app, in light and dark, and in English or Russian. The scheduled task form no longer shows a different widget on every browser, and the timezone box in the admin AI tab suggests common zones in a styled list while still letting you type any zone.

462 words · patch release

Read post
v2.6.212026-10-12

Model picker and Incognito notice stay in place

The model picker in the chat composer now opens lined up with the composer instead of sliding off the right edge of a phone screen, and the Incognito notice sits centered above the composer.

387 words · patch release

Read post
v2.6.202026-10-10

Incognito moves to the top right

The Incognito button now lives in the top right corner of the chat, next to Upgrade and Share, which is where you would look for it in other chat apps. It used to sit on the left of the header beside the menu. It is a quiet icon with a label, and on phones the label drops away so only the icon shows, which keeps the header from getting crowded. When incognito is on, the button lights up in the accent color so you can tell at a glance that the chat is not being saved. Nothing about how incognito works has changed: an incognito chat is still never saved to your account or this browser, it still stays out of your memory and learned preferences, it still can't be shared by link, and it still disappears when you leave it. The button only appears on chats that are not part of a project, and it locks once an incognito chat has messages, so you can't flip a chat that may already have been saved. This is purely a change of position and styling, made so the feature is easy to find the moment you need it. Credits are still used in incognito chats because the models cost the same to run, and the in-chat explanation of what incognito does is unchanged in both English and Russian.

588 words · patch release

Read post
v2.6.192026-10-10

Incognito chats

Sometimes you want to ask something without it following you around. Incognito chats are a new way to do that: pick the Incognito button at the top of an empty chat and everything you say in it stays out of your history. The conversation is never saved to your account or to this browser, it doesn't appear in your chat list, and it is gone the moment you leave it or reload the page. Portail also keeps your Enhanced Context memory and your learned preferences out of the picture in both directions. Nothing from your saved memory is used to answer you, and nothing you say is added to it or used to tune how Portail talks to you later. The chat can't be shared as a link either, since a shared link would be a saved copy. Two things stay the same on purpose: credits are still used for every reply, because the models cost the same to run, and a chat that already has messages can't be switched into incognito halfway through, since it may already have been saved. To start one, open an empty chat and press Incognito, or press it from anywhere to begin a fresh incognito chat. The button works the same on phones, and the explanation of what incognito does shows right in the empty chat so you never have to guess. Incognito is available in English and Russian.

612 words · patch release

Read post
v2.6.182026-10-10

Friendlier sign-in errors, a keyboard-friendly reset window, and one Portail address

A handful of small fixes around logging in. First, if you type www.portail.cc, you now land on portail.cc, with the page and any link details kept. The www address used to load the full site on its own, but our sign-in provider only trusts portail.cc, so Google sign-in failed there and password reset links lost their way back. Now there is one address and everything on it works. Second, the login form, the sign-up page and the email code check no longer show raw technical errors. Messages that started with the word Firebase and ended in a code in brackets are gone. If you try too many times in a row, you see a plain note asking you to wait a few minutes. If your connection drops, you are told to check it and try again. If an account has been disabled, you are told so and pointed to support. Anything we haven't written a specific message for falls back to a simple "Something went wrong, please try again" instead of a code. All of this is in English and Russian. Third, the "Forgot password?" window now behaves properly with a keyboard and a screen reader. It opens with the cursor already in the email field, Tab and Shift+Tab stay inside the window instead of wandering off to the page behind it, Escape closes it, and when it closes your focus goes back to the "Forgot password?" link. Screen readers announce it as a dialog with its title. Nothing about your account or plan changes.

638 words · patch release

Read post
v2.6.172026-10-10

Password reset emails work again

Asking for a new password from the login page had two separate problems. The first one you could see: sometimes the request failed with a technical error saying a domain was not allowlisted. That happens when the address you opened Portail from is not on our sign-in provider's list of approved sites, so it refused to build a reset link that points back to it. Now, when that happens, we still send you the reset email, just without the Continue button that takes you back to Portail at the end, and you no longer see a raw error with the word Firebase in it. The second problem was quieter. Even when the email did go out, the link in it brought you back to the login page without ever letting you pick a new password, because we had told the provider that Portail would finish the reset itself and no page on our side was doing that. Now the link opens a simple page where you type your new password, and once it is saved a Continue button brings you back to Portail so you can log in with it. Nothing changes about your account, your saved chats or your plan, and passwords you already use keep working. If you requested a reset recently and the link did not help, request a fresh one and it will take you to the new page. Links from older emails may already have expired, which is normal for reset links.

621 words · patch release

Read post
v2.6.162026-10-10

The password reset window now dims the whole page

If you opened "Forgot password?" on the log in page, the reset window appeared, but the dark backdrop behind it only covered a narrow strip the width of the log in card. The rest of the page stayed bright, so it looked like two layers fighting over the screen and it was hard to tell what you could still click. The cause was a small layout detail: the log in card slides up into place when the page loads, and the leftover position from that animation quietly shrank the area the backdrop was allowed to fill. The reset window now opens on top of the whole page instead of inside the card, so the backdrop dims everything edge to edge, on a phone and on a wide monitor alike. Nothing else about the flow changed. You still type your email, press Send reset link, and we email you a link to choose a new password. Clicking anywhere on the dimmed area or pressing Back to log in closes the window and takes you back to the form exactly as you left it, with whatever you had typed still there. This is a visual fix only. Your account, your password and the way reset emails are sent are untouched, and there is nothing you need to do on your side. If the page still looks off after this update, a hard refresh will pick up the new version.

624 words · patch release

Read post
v2.6.152026-10-10

One composer, one focus

The place you type got simpler. A new chat now shows one question and the composer, nothing else competing for your eye; your activity grid moved to Profile, under General. The composer keeps the same rounded shape whether the chat is empty or full, with bigger text that no longer makes phones zoom in when you tap it. The model name sits there as plain text, and the cost hint lives in the model list where you actually compare models. The round button on the right does one job at a time: with an empty box it starts a voice conversation, and as soon as you type or attach something it becomes send. In a running chat the composer floats over the conversation with a soft fade instead of sitting under a divider line. The box on the home page and the demo below it got the same composer, so what you see before signing up is what you get after.

545 words · patch release

Read post
v2.6.142026-10-10

A calmer chat that puts the answer first

The chat got quieter so the conversation can do the talking. The page is now a soft charcoal instead of near-black, and things that sit on it, like your messages and the composer, are one step lighter, so you can tell what's what without borders everywhere. Answers are written in the main text color instead of a secondary gray, which makes long replies much easier to read. The sidebar is one simple list: New chat, Canvas and Ratchet at the top, then your projects and chats with plain, quiet labels and tighter rows, so more of your history fits on screen. The Projects pages now share the exact same sidebar, so moving between a project and a chat no longer changes the look under you. Light mode follows the same idea with an off-white sidebar next to a white page. The theme switch moved out of the sidebar footer; it lives in Profile under Customization, next to language.

540 words · patch release

Read post
v2.6.132026-10-09

Proper home screen icon on iPhone and iPad

Adding Portail to your iOS home screen used to give you a plain letter tile instead of our icon. It now uses a dedicated, correctly sized icon, so the app looks like an app.

385 words · patch release

Read post
v2.6.122026-10-09

A clear cookie policy and your choice of analytics

Portail now explains exactly what it stores in your browser, and puts optional analytics behind your say-so. A short banner asks whether we may count anonymous sign-up steps to improve the product, with equal-weight Only necessary and Accept analytics buttons. Until you answer, nothing optional is stored or sent. You can change your mind at any time from Cookie settings in the footer. A new Cookie Policy page, in English and Russian, lists everything we store and why: your sign-in session and preferences, the small cookie that sends signed-in visitors straight to their chat, and the anti-abuse protection that keeps free allowances fair. The Privacy Policy has a matching new section. We use no advertising cookies and no third-party trackers.

499 words · patch release

Read post
v2.6.112026-10-09

Signed in? Straight to your chat

If you already have a Portail account and open portail.cc, you no longer land on the marketing page and then have to click through. The root address now sends you directly to your chat before the page even draws, so there is no flash of landing copy and no extra click. Visitors without an account see the landing page exactly as before, and search engines still read it normally. Signing out brings the landing page back for that browser.

440 words · patch release

Read post
v2.6.102026-10-09

Your question makes it to the chat

The ask box on the home page now reliably gets your question into the chat. Before, a question typed there could be dropped on the way, or land in the wrong place if you were already signed in. Now it always goes into a fresh chat, on Auto or the model you picked, and your older conversations stay untouched. The send button also answers the moment you press it, so a slow connection no longer looks like nothing happened.

501 words · patch release

Read post
v2.6.92026-10-092 screenshots

A new front door

The home page at portail.cc is rebuilt around one idea: GPT, Claude and Gemini in one chat, for one $20 plan instead of three. The top of the page is now the same input box you use in the app, with the real model picker inside it. Type a question, sign up, and it sends itself as your first message on the model you picked. Below it, a scripted demo plays real Portail screens: switching models mid-thread, Gmail and Calendar actions, a scheduled morning brief, a Council of models, and Canvas.

641 words · patch release

Read post
v2.6.82026-10-09

Sign up first, questions after

Getting into Portail used to mean answering seven or more questions before you could even make an account, and if your phone reloaded the page while you fetched the email code, you landed back on an empty form. Now you give your name and age, make the account, and only then answer the short questionnaire that tunes the chat to you. The email code screen survives a reload, the sign-up screen tells you up front when Google sign-in can't work inside an app's built-in browser, and the pricing table opens on monthly prices so the Pro trial is in plain sight. We also fixed the annual discount label: it said about 17%, the real saving is about 4%.

632 words · patch release

Read post
v2.6.72026-10-08

Your calendar, in your time zone

Connected Google Calendar now shows what is actually on it, and events land at the local time you meant instead of UTC. Gmail can look over several emails at once and read a whole message when you need to answer it.

428 words · patch release

Read post
v2.6.62026-10-08

Scroll stays where you put it

Scrolling up while Portail is still writing a reply now holds your place with a trackpad, a touch screen or the scrollbar. Drag or scroll back to the bottom and the chat picks the reply up again.

396 words · patch release

Read post
v2.6.52026-10-07

A rebuilt admin panel behind the scenes

Nothing changes in your chats this time. We rebuilt the panel our team uses to run Portail, so revenue figures count legacy plans correctly, account fixes save on the first try, and support can help you faster. Subscribers on the older Ultra++ plan now see its proper name instead of Ultra Plus Plus.

425 words · patch release

Read post
v2.6.42026-10-07

Scroll back while a reply is still writing

You can now scroll up through a conversation while Portail is still typing its answer. The chat stays where you left it instead of dragging you back to the bottom, and once you scroll down to the end it follows the reply again.

413 words · patch release

Read post
v2.6.32026-10-07

A quieter savings panel

The Plan Fit section in Usage & Billing loses its fine-print footnote, so the panel ends on the savings figure, your last 30 days and the plan verdict.

372 words · patch release

Read post
v2.6.22026-10-07

Your savings, in one number

The Plan Fit section in Usage & Billing now leads with one figure: what Portail saves you against buying the same AI labs' own plans separately, measured against what you actually pay. The 30-day summary is three numbers and a line, and the plan verdict is one sentence.

428 words · patch release

Read post
v2.6.12026-10-06

A quiet activity grid on the empty chat

The empty chat screen now shows a GitHub-style grid of how often you have chatted over the last 30 weeks, right under the input box.

384 words · patch release

Read post
v2.6.02026-10-062 screenshots

Know your pace

Portail now shows you your own usage: your last 30 days in Usage & Billing, roughly how many messages are left before a free limit, and which plan fits the way you actually use it.

691 words · minor release

Read post
v2.5.22026-10-05

Telegram inline asks cost less

Typing @ the Portail bot in any Telegram chat now offers a plain Ask AI result and answers on a fast, low-cost model, whatever model your private chat uses. Your private chat keeps the model you picked.

397 words · patch release

Read post
v2.5.12026-10-05

Fixes for images, voice, gifts and sign-in

A round of fixes found in a full review of Portail. Chats that start with an image now keep their title, voice mode stops reading hidden markers out loud, a gift link can only be redeemed once, and signing in can no longer bounce you to another website. Behind the scenes, closing a tab now stops the work it started, and credit balances stay correct when two requests land at the same moment.

535 words · patch release

Read post
v2.5.02026-10-032 screenshots

API keys for your own code

Paid plans now include API keys. Create one in Settings > Developer API and your own server can reach your Portail account over HTTPS: search what Portail remembers about you, read your connected apps, and act through them, like sending an email from your Gmail or adding an event to your calendar. The API never runs a model, so calls don't spend credits. Every key carries only the permissions you pick, writes need an idempotency key so a retry can't send the same email twice, and each account has a daily cap on connector calls. Full docs, with curl examples, are at portail.cc/developers.

815 words · minor release

Read post
v2.4.52026-10-032 screenshots

A front page that shows instead of tells

The portail.cc front page now shows the product instead of describing it. Right under the headline, a short chat plays out in one thread: GPT-5.5 drafts a launch tweet, Claude rewrites it calmer, and Gemini translates it to French. Nobody re-explains anything between turns, which is the whole point of Portail, and now you can see it in about ten seconds instead of reading three sections to get there. The page also asks for less. There is one button at the top, Start free, with the plan spelled out underneath: free plan, no card, every model. On phones, where the floating model logos are hidden, the top of the page used to be a headline on an empty screen. Now the chat sits right under it. We also removed the comparison table that lined Portail up against Typing Mind, T3.chat and Poe. A grid of checkmarks we wrote about ourselves was never the most convincing part of the page, and the demo makes the same case better.

609 words · patch release

Read post
v2.4.42026-10-03

Ask Portail from any Telegram chat

Type @ and the Portail bot's name, then a question, in any Telegram chat, and the answer appears right there for everyone in it. It runs on your account and credits, and every answer is also saved in your Portail chats.

420 words · patch release

Read post
v2.4.32026-10-03

Portail now asks your age

Portail now asks how old you are when you sign up, on the web and in Telegram. You type a number, and we keep only an age range, never the exact age. The reason is simple: we can't knowingly hold chats from kids under 13 without a parent's consent, and we have no way to collect that consent. So under 13 means no account, the same rule OpenAI uses. If you're 13 to 17, you can use Portail with a parent's permission, and you get a short game first that covers five rules for using AI well: double-check facts that matter, keep personal info out of chats, use AI to learn instead of skipping the learning, take real problems to real people, and never make fake pictures of real people. Your chats also get a little extra care around topics that matter more at that age. If you already had an account before this change, you'll see a one-time question the next time you open the chat. Answer it once and it never comes back. Our Terms of Service now spell out the 13+ rule in both English and Russian.

660 words · patch release

Read post
v2.4.22026-10-02

Portail Code is now Ratchet

The coding companion to Portail is now Ratchet, a small coding agent that runs in your terminal. Link it to your Portail account once with ratchet --link and it runs on your plan's credits, with no API key to paste. Chats that turn into something buildable can still be handed off: Ratchet picks up the build brief on your computer, builds it, and reports back when it's done. code.portail.cc is now Ratchet's home, with install steps for Linux and macOS.

536 words · patch release

Read post
v2.4.12026-10-01

Rename your chats, and a Canvas that fits your phone

You can now rename any saved chat from the sidebar, and Canvas has a new mobile layout with a bottom prompt bar and an options sheet.

414 words · patch release

Read post
v2.4.02026-09-302 screenshots

Chat replies tuned to how you work

When you signed up, Portail asked what you do, what you're mainly there for, and how familiar you already are with AI assistants. Those answers used to go nowhere, sitting in the database and only ever showing up in the admin panel. Every chat reply now reads them first. A developer debugging something gets a more direct answer with less hand holding. Someone new to AI assistants gets plainer language instead of jargon. None of this shows up as a new setting or a new screen. It works quietly, on every account that already finished onboarding, starting with this release.

701 words · minor release

Read post
v2.3.42026-09-282 screenshots

New pricing: Pro, Plus, and Ultra

Standard, Pro, Ultra, and Ultra++ are replaced by three plans: Pro ($20/mo), Plus ($80/mo), and Ultra ($250/mo). Each plan's monthly credit allowance is now sized against what the underlying models actually cost to run, not a number picked early on. Existing subscribers keep their current plan, price, and credits exactly as they are; nothing changes for you unless you choose to move to the new lineup yourself. New signups see the new pricing starting now.

466 words · patch release

Read post
v2.3.32026-09-282 screenshots

Gift cards get a Digital PDF option

Printable gift cards had one export path: a fold-and-cut sheet with the back face printed upside down so it reads right side up once folded. That made sense on paper but broke anyone trying to save the card as a normal PDF, since the back page came out inverted with no fold to fix it. The gift card page now offers a Format choice next to the print button: Fold card keeps the original single-sheet layout for printing and folding by hand, and the new Digital PDF option lays the front and back out as two separate upright pages with a page break between them, so saving to PDF from the print dialog produces a clean two-page file in the right order.

488 words · patch release

Read post
v2.3.22026-09-281 screenshots

Drag files straight into the chat

The chat composer now accepts drag-and-drop: drop a file or screenshot anywhere on the chat window and it lands as an attachment, no need to hunt for the paperclip button first.

406 words · patch release

Read post
v2.3.12026-09-282 screenshots

Gift cards you can actually print

Every Portail gift now comes with a card worth handing over. The art on it is drawn from the gift's own code, so no two cards look the same. Pick a style, a light or dark tone, and the recipient's name, then print it on any home printer. One sheet, single-sided, folds into a two-sided card with a QR code on the back that takes them straight to redemption.

503 words · patch release

Read post
v2.3.02026-09-262 screenshots

Scheduled tasks: Portail works while you're away

Write a prompt once, pick when it runs, and Portail runs it on its own. A weekday briefing from your Gmail and Calendar at 8:00, a Friday digest of competitor news, a one-off reminder next Tuesday. Each task keeps its own chat thread, and the answer is delivered to Telegram, your inbox, or just the web. Reply in Telegram and you pick up that same thread. Tasks that would change something in a connected app, like sending an email or adding a calendar event, stay read-only until you turn writes on for that task, and every action a run takes is listed at the top of its result.

749 words · minor release

Read post
v2.2.62026-09-261 screenshots

Clean Canvas images on Ultra

Ultra and Ultra++ members can now switch off the Portail watermark in Canvas and download clean images straight from the gallery. The new Watermark switch sits in the Canvas form above the model list and stays on by default, so nothing changes unless you flip it. Pro keeps the mark on every image and sees the switch locked with a pointer to Ultra.

422 words · patch release

Read post
v2.2.52026-09-261 screenshots

Inline code stays inline

Short bits of code inside a sentence, like a file name or a command, show up as small highlighted words again instead of breaking the reply into a stack of code boxes. Real code blocks still get their own card with a Copy button.

388 words · patch release

Read post
v2.2.42026-09-261 screenshots

Quieter thinking indicator

The thinking label above a reply no longer sits inside a bordered box. While the model reasons you see a shimmering Thinking label, and clicking it unfolds the raw reasoning right under it, no panel around it.

385 words · patch release

Read post
v2.2.32026-09-241 screenshots

Better answers for resumes, PRDs, and a few other specialties

Portail's chat model now has access to a small library of domain playbooks it can quietly pull in mid-reply: resume and CV formatting, cover letters, content and copywriting frameworks, product management frameworks like RICE scoring and PRD structure, presenting data to stakeholders, and spreadsheet formulas. The model only sees a short list of what is available up front. It fetches the actual guidance only when a request calls for it, the same on-demand technique already used for personal memory and project files, so the improvement costs nothing on ordinary turns that never touch these topics. The lookup and the guidance it returns are never shown to the user, only a more informed answer.

511 words · patch release

Read post
v2.2.22026-09-231 screenshots

Chat replies no longer cut off mid-answer on slow turns

A long reasoning phase or a slow tool round-trip could leave the connection between the browser and Portail's server silent long enough for a proxy in between to treat it as idle and close it. That looked exactly like the model finishing its answer early, with no error shown, so a reply could just stop mid-sentence for no visible reason. The chat stream now sends a small keep-alive signal every 15 seconds while it waits on the model, and an aborted reply is now logged with how much text had already streamed, so a genuine drop is distinguishable from a deliberate stop.

485 words · patch release

Read post
v2.2.12026-09-231 screenshots

Live status for actions, not just text

Chat used to go quiet while the model checked your email or calendar, then pick back up like nothing happened. Every action it takes now shows up as it happens, styled distinctly from the reply itself, and each one reports how long it took once it's done.

425 words · patch release

Read post
v2.2.02026-09-221 screenshots

Sheets, Docs, Tasks, and Instagram join the Connectors list

Portail's chat can now act on four more accounts: Google Sheets, Google Docs, Google Tasks, and Instagram. Connect any of them from Settings > Connectors and the assistant can create a spreadsheet and read a range back, spin up a Google Doc and drop a paragraph into it, check off a task without opening the Tasks app, or post an image to a connected Instagram business account and answer a comment. Nine connectors now share the same managed sign-in as the five that shipped before them, so linking one from the main chat or from Portail Agents links it in both.

732 words · minor release

Read post
v2.1.12026-09-221 screenshots

Appearance and language settings, plus a calendar connector that can list events

Settings > Customization now has its own Appearance and Language sections instead of leaving those controls scattered around the app. Dark and Light are now a visible, labeled choice instead of a single footer icon, and English and Russian can be switched right from inside the chat app, not just before signing in. The Google Calendar connector also grew a new tool: it can list upcoming events, so the assistant can check what is on your calendar without first needing to create or remove anything.

489 words · patch release

Read post
v2.0.22026-09-211 screenshots

Gift purchases no longer get stuck on Payment processing

Gift subscriptions could get stuck showing Payment processing indefinitely when the payment webhook was delayed or never arrived, even though the giver's card had already been charged. Portail now verifies a pending gift's payment directly with the payment provider as a fallback, the same self-healing check regular subscription purchases already had, so a delayed or missing webhook no longer leaves a paid gift stranded on pending.

434 words · patch release

Read post
v2.0.12026-09-181 screenshots

Smoother chat scrolling and a fixed settings modal

Three chat interface bugs are fixed. Message rows no longer freeze at a placeholder height once they scroll out of view, which was corrupting the scroll container and causing sudden jumps. Auto-scroll no longer mistakes a layout shift for the user scrolling back down during a streaming reply, so scrolling up to read something now reliably pauses it. The account settings modal now locks background scroll while it's open and no longer clips its own content in a narrow flex layout.

480 words · patch release

Read post
v2.0.02026-09-164 screenshots

Connectors, gifted subscriptions, and a live look at how the model thinks

The chat can now reach outside the conversation: link Google Calendar and Gmail under Settings > Connectors and it can check your day, draft a reply, or find a slot without you leaving the thread. Composio handles the sign-in, so no OAuth token ever touches Portail's database. This release also adds gifted subscriptions (pay once, the recipient gets a code by email, no account needed until they redeem it), a live view of a model's reasoning as it streams in instead of a blank line while it thinks, and a quote-reply button that appears when you select text in a reply.

1251 words · major release

Read post
v1.8.32026-09-131 screenshots

Project pages now match the Projects sidebar

Opening a single project used to drop you into a page with just a logo and a back link. It now opens inside the same app sidebar as the Projects list and chat, with Chat and Canvas at the top, links to Code and Agents, the open project highlighted in the Projects list, your other chats below, and your account and theme control at the bottom.

458 words · patch release

Read post
v1.8.22026-09-121 screenshots

Projects moves into the same sidebar as chat

The Projects page now uses Portail's app sidebar instead of a separate header. Projects, regular chats, Code, Agents, Canvas, theme controls, and account identity stay within reach while you organise your work, and the duplicate Portail logo and Back to chat button are gone from the page header.

313 words · patch release

Read post
v1.8.12026-09-122 screenshots

Auto mode picks a model on merit, and image previews open full size

Auto mode was quietly sending almost every message to the cheapest model in the lineup, no matter how hard the question was. It now makes a real routing decision per message, so a stack trace reaches a coding model and a proof reaches a reasoning model while small talk still goes somewhere fast and cheap. Image previews also stopped opening in a cramped, clipped box, and the Projects page got a cleaner layout.

439 words · patch release

Read post
v1.8.02026-09-111 screenshots

Projects: your instructions and files, attached to the work

A project is a named container holding custom instructions and reference files, and every chat you start inside it follows them. Set one up for anything you come back to, and stop re-explaining the same context at the top of every conversation. Your three most recent projects now sit above the chat list in the sidebar, with the rest on a new Projects page.

719 words · minor release

Read post
v1.7.42026-09-101 screenshots

A new chat opens with a different line every time

The empty chat screen used to greet everyone with the same fixed line, every single time. It now draws from a pool of twenty different openers in English and Russian, each written as its own direct invitation rather than a generic placeholder, so starting a new conversation doesn't feel like reading the same header for the thousandth time.

329 words · patch release

Read post
v1.7.32026-09-101 screenshots

Portail Code comes to Linux, and the model switcher stops hiding itself

The Portail Code marketing site now offers a Linux AppImage download alongside the Windows installer, served through the same checksum-verified route. Separately, the model switcher's dropdown could render partly off-screen once the search field grabbed focus, because it sized itself against a static viewport unit that a mobile keyboard does not shrink. It now measures the space actually available and adapts to it.

422 words · patch release

Read post
v1.7.22026-09-101 screenshots

Portail SH becomes Portail Code, with a steadier editor underneath

The desktop coding harness is now Portail Code everywhere it appears. This release also fixes several editor problems that could lose a handoff, leave stale tabs open, hide useful errors, or make light mode feel unfinished. Existing Portail SH settings and conversations move forward automatically, and the Windows download is now pinned to the exact v1.7.2 installer with a verified checksum.

393 words · patch release

Read post
v1.7.12026-09-101 screenshots

Free accounts sharing one browser now share one allowance

Signing up for a fresh free account from the same browser or device used to reset the free credit pool from zero every time. Free accounts that repeat a strong signal, a random installation id, a verified phone or email alias, or a Telegram id, now draw from one shared allowance instead. A separate gap in account creation was also closed: the captcha check ran as a standalone call that scripted signups could skip entirely, so it now runs as a single-use, action-scoped token inside the signup request itself.

530 words · patch release

Read post
v1.7.02026-09-062 screenshots

A new home page lineup and a door marked With Portail

The landing page headline now leads with the model count instead of a slogan, the model showcase grid is generated straight from the live model list instead of a hardcoded lineup, and a new experiments microsite at with.portail.cc gives early tools and partner pilots a public home before they graduate into the main app.

671 words · minor release

Read post
v1.6.32026-09-061 screenshots

Office and PDF uploads stop getting silently rejected

Attaching a PDF or Word document to a chat could fail before it ever reached the model, because the upload filter required the browser's reported file type to match its extension, and browsers are unreliable about that for office documents. The extension alone now decides, which is what actually controls how the file is served back. AI-generated file downloads also stopped opening the file inline instead of downloading it.

439 words · patch release

Read post
v1.6.22026-09-041 screenshots

The weekly persona tuner stops choking on its own replies

The reinforcement job that proposes small weekly wording edits to the assistant's persona now gives its model enough room to think before answering, instead of failing with an invalid-JSON error.

405 words · patch release

Read post
v1.6.12026-08-292 screenshots

The release archive is public

Portail now has a public version log with indexed changelog posts, screenshots, and deploy checks that block incomplete release notes.

321 words · patch release

Read post
v1.6.02026-08-292 screenshots

Memory makes more room

Enhanced Context now compresses older passive memories while preserving recent details, which gives long-running chats more useful context room.

646 words · minor release

Read post
v1.5.52026-08-251 screenshots

Long chat streaming is faster

Long conversations now stream more smoothly, reducing slowdown when large histories are present.

367 words · patch release

Read post
v1.5.42026-08-251 screenshots

Memory baseline patch

The memory baseline was corrected so stored context starts from a cleaner, more predictable state.

362 words · patch release

Read post
v1.5.32026-08-251 screenshots

Link admin failures are visible

The link admin now shows load failures instead of leaving the operator with an unclear empty state.

370 words · patch release

Read post
v1.5.22026-08-251 screenshots

Link admin auth failures stay local

Authentication errors in the link admin are handled locally so failures do not spill into unrelated admin surfaces.

371 words · patch release

Read post
v1.5.12026-08-251 screenshots

Admin request cap removed

The admin panel request cap was removed so internal operations are not blocked by an artificial ceiling.

367 words · patch release

Read post
v1.5.02026-08-251 screenshots

Fairer usage limits

Usage limits were reworked around monthly, weekly, and session windows so credit pressure is easier to understand.

575 words · minor release

Read post
v1.4.182026-08-251 screenshots

Russian coverage is complete

Remaining chat, Canvas, dataroom, connect, and plan copy was translated so Russian users see fewer English fallbacks.

367 words · patch release

Read post
v1.4.172026-08-231 screenshots

Text fences render normally

Plain text code fences now render through the normal markdown path instead of a special passthrough box.

366 words · patch release

Read post
v1.4.162026-08-231 screenshots

Auto and Council become switches

Auto routing and Council comparison controls now sit as switches under the model picker.

363 words · patch release

Read post
v1.4.152026-08-231 screenshots

Auto routing and Council mode

Portail added automatic model routing and Council comparison mode for multi-model answers.

364 words · patch release

Read post
v1.4.142026-08-231 screenshots

Sharing polish and admin memory

Conversation sharing gained social targets, share analytics, admin remember-me, and model-picker fixes.

365 words · patch release

Read post
v1.4.132026-08-231 screenshots

Conversation sharing

Users can create shareable conversation links from Portail chats.

358 words · patch release

Read post
v1.4.122026-08-211 screenshots

Continual learning

Portail added per-user style adaptation and weekly platform reinforcement so responses can better match a user's habits.

362 words · patch release

Read post
v1.4.112026-08-201 screenshots

Legal pages match the product

The Terms, Privacy Policy, and Refund Policy were updated for Agents, Canvas, Telegram, and OpenRouter.

364 words · patch release

Read post
v1.4.102026-08-201 screenshots

Light theme model icons

Model icons were corrected for the light theme so provider marks stay legible.

363 words · patch release

Read post
v1.4.92026-08-201 screenshots

Cleaner model list and clearer privacy

OpenRouter meta-routing models were hidden from the chat picker, and privacy copy now explains model-training data use.

374 words · patch release

Read post
v1.4.82026-08-201 screenshots

Canvas image watermark

Canvas-generated images now receive a translucent Portail mark.

359 words · patch release

Read post
v1.4.72026-08-201 screenshots

Mobile touch fixes

Mobile navigation and touch targets were fixed across the client, Codeweb, and Agents surfaces.

361 words · patch release

Read post
v1.4.62026-08-201 screenshots

Account creation waits for email

Signup now defers account creation until email verification is complete.

360 words · patch release

Read post
v1.4.52026-08-191 screenshots

Real spend in admin

The admin panel now shows real OpenRouter spend and usage, including a corrected Random Stuff spend figure.

365 words · patch release

Read post
v1.4.42026-08-191 screenshots

Canvas links back to chat

Canvas gained a back-to-chat link so users can move between the visual workspace and the conversation.

369 words · patch release

Read post
v1.4.32026-08-191 screenshots

Streaming scroll and document output fixes

Chat auto-scroll no longer fights user scrolling, markdown tables look better, and generated documents support tables and links.

371 words · patch release

Read post
v1.4.22026-08-181 screenshots

Canvas wordmark in the sidebar

The sidebar now shows the Portail Canvas wordmark.

363 words · patch release

Read post
v1.4.12026-08-181 screenshots

Ultra++ and billing lifecycle fixes

Ultra++ was added alongside immediate cancellation and billing lifecycle corrections.

357 words · patch release

Read post
v1.4.02026-08-171 screenshots

Canvas, billing, and agent polish

The 1.4 line bundled Canvas naming, billing improvements, image-generation updates, agent landing work, and model sync tooling.

575 words · minor release

Read post
v1.3.12026-08-171 screenshots

Canvas gets a sidebar home

The old Agentic entry became Canvas, with a visual workspace placeholder in the sidebar.

365 words · patch release

Read post
v1.3.02026-08-141 screenshots

Safer uploads and sign-in hardening

Portail hardened uploads, sign-in, and desktop isolation for the v1.3.0 release.

563 words · minor release

Read post
v1.1.32026-08-141 screenshots

Keyboard navigation and model picker

Keyboard navigation and the model picker were tightened for faster movement through the chat UI.

361 words · patch release

Read post
v1.1.22026-08-141 screenshots

Image generation lands

Image-capable models gained a dedicated image action and image prompts route through the correct generation path.

572 words · minor release

Read post
v1.0.22026-08-131 screenshots

Fresh starts and fair timing

The chat start flow and timing behavior were corrected after the first 1.0 releases.

361 words · patch release

Read post
v1.0.12026-08-101 screenshots

Sharper copy and cleaner product framing

Public copy was tightened and vendor names were reduced where they distracted from the Portail product.

363 words · patch release

Read post
v1.0.02026-08-091 screenshots

Every model, one allowance

Portail shipped the full OpenRouter catalog with credit-only model access.

1057 words · major release

Read post
v0.4.22026-08-091 screenshots

Email verification PIN

Email verification now uses a 6-digit PIN during signup.

354 words · patch release

Read post
v0.4.12026-08-081 screenshots

Registration captcha

Turnstile captcha was added to registration to reduce abusive signups.

356 words · patch release

Read post
v0.4.02026-08-071 screenshots

UI and UX upgrades

The client received a wider UI and UX pass across chat, pricing, navigation, and comparison surfaces.

572 words · minor release

Read post
v0.3.212026-08-031 screenshots

Omniverse on Telegram

Portail reached Telegram through Omniverse, sharing product accounts, conversations, credits, and memory.

362 words · patch release

Read post
v0.3.202026-08-011 screenshots

Product tour

A product tour was added so new users can understand core chat controls after release notes.

364 words · patch release

Read post
v0.3.192026-08-011 screenshots

Limited-time offer

Existing users received a limited-time 13 percent offer through a gated promotion banner.

366 words · patch release

Read post
v0.3.182026-08-011 screenshots

Passive memory lookup

Passive memory moved to on-demand lookup instead of being sent on every turn.

362 words · patch release

Read post
v0.3.172026-07-301 screenshots

Agents security and billing fixes

Portail Agents received security and billing fixes around subscription gates and agent behavior.

365 words · patch release

Read post
v0.3.162026-07-301 screenshots

Daily feature-tip email

Daily feature-tip email support was added with localized copy and controlled sending.

360 words · patch release

Read post
v0.3.152026-07-291 screenshots

Fairer metering and safer storage

Token metering, credit leaks, authentication, and storage paths were hardened.

356 words · patch release

Read post
v0.3.142026-07-261 screenshots

Every model through OpenRouter

The model stack moved to OpenRouter as the single provider for the full chat lineup.

365 words · patch release

Read post
v0.3.132026-07-241 screenshots

Voice mode stall fixes

Voice mode no longer hangs on stalled model output, and provider stalls are watched more carefully.

366 words · patch release

Read post
v0.3.122026-07-231 screenshots

Required email verification

Signup now requires email verification before users enter the main app.

361 words · patch release

Read post
v0.3.112026-07-221 screenshots

Edit sent messages

Users can edit previously sent messages and continue from the corrected prompt.

364 words · patch release

Read post
v0.3.102026-07-221 screenshots

Annual billing and Pro trial changes

Annual billing landed, and the Pro trial moved to monthly-only behavior.

360 words · patch release

Read post
v0.3.92026-07-201 screenshots

CORS and title timeout fixes

CORS credential handling and chat title timeout behavior were fixed.

357 words · patch release

Read post
v0.3.82026-07-201 screenshots

PDF and document uploads are readable

PDF and document uploads were fixed so selected models can read their extracted content.

363 words · patch release

Read post
v0.3.72026-07-201 screenshots

Admin email broadcast

Admins can send email broadcasts to all users or selected users.

358 words · patch release

Read post
v0.3.62026-07-191 screenshots

In-chat image preview

Image attachments can be previewed directly in chat.

358 words · patch release

Read post
v0.3.52026-07-171 screenshots

SEO and answer-engine refresh

Public SEO, answer-engine metadata, model-lineup text, Canvas discovery, and status metadata were refreshed.

360 words · patch release

Read post
v0.3.42026-07-171 screenshots

Voice mode announcement

Voice mode was announced through the release modal after landing in the composer.

365 words · patch release

Read post
v0.3.32026-07-171 screenshots

Generated-file downloads and streaming stalls

Generated file downloads and streaming stall behavior were fixed.

355 words · patch release

Read post
v0.3.22026-07-141 screenshots

Model watchdog and status page

A model watchdog and live status page were added during the early provider-health phase.

364 words · patch release

Read post
v0.3.12026-07-121 screenshots

Local time awareness

Chat gained quiet awareness of the user's local time.

355 words · patch release

Read post
v0.3.02026-07-111 screenshots

Files become model-readable

Models can read uploaded files, and the first Canvas groundwork entered the codebase.

570 words · minor release

Read post
v0.2.72026-07-091 screenshots

New model lineup

The model lineup was refreshed with GPT-5.5, Claude Sonnet 5, Gemini 3.1 Pro, and GLM-5.2.

362 words · patch release

Read post
v0.2.62026-07-081 screenshots

Memory recovery and hardened prompts

Memory recovery, the contextual desktop bridge button, and prompt hardening shipped together.

360 words · patch release

Read post
v0.2.52026-07-081 screenshots

SEO and AEO foundation

The first SEO and answer-engine layer landed for portail.cc and code.portail.cc.

358 words · patch release

Read post
v0.2.42026-07-061 screenshots

Connect updates and Codeweb

Connect updates shipped alongside the standalone Codeweb app inside the deployable repository.

363 words · patch release

Read post
v0.2.02026-07-031 screenshots

Credit billing fixes

Credit billing was corrected, localhost fallbacks were removed, and composer newline behavior was fixed.

565 words · minor release

Read post
v0.1.52026-07-031 screenshots

Design updates

The early product design was tightened after the first billing and linking passes.

357 words · patch release

Read post
v0.1.42026-07-021 screenshots

Early release cleanup

The v0.1.4 patch gathered early fixes before the credit-billing line began.

357 words · patch release

Read post
v0.1.32026-06-301 screenshots

Linking fixes

Early account and device linking work was corrected before the next public patch.

356 words · patch release

Read post
v0.1.22026-06-271 screenshots

Qwen thinking blocks collapse by default

Qwen thinking blocks now collapse by default so chat answers stay readable.

364 words · patch release

Read post
v0.1.12026-06-271 screenshots

Scroll fixes

Early scroll behavior was patched so chat navigation felt less jumpy.

354 words · patch release

Read post
v0.1.02026-06-171 screenshots

MVP

The first Portail MVP shipped with the core chat experience and the foundation for real-time synced conversations.

572 words · minor release

Read post