Every feature, improvement, and fix as we ship it — from the team building the AI support agent that resolves, not deflects.
Asks is officially live, and we want to remember who showed up first. The first 50 stores to subscribe — on Shopify or anywhere else — get 50% off their first 3 months after the 7-day free trial. Any plan, monthly or yearly.
Use code FOUNDING50at the checkout step in your dashboard's Subscription page. On yearly plans the same value is applied as a discount on your first year. Once 50 stores have redeemed it, it's gone.
Real customers don't compose essays. They type “hey”, then “is the canvas weekender back in stock”, then “the olive one” — three messages in as many seconds. The agent now treats a burst like that as one thought: every reply considers every message that hasn't been answered yet, attachments included, so a photo followed by “do you have this in olive?” keeps the photo in the picture.
And when a new message lands while the agent is mid-reply, the in-flight answer steps aside quietly instead of racing the new one — you never get two replies fighting over the same question. Escalation and sentiment detection read the full burst too.
The widget's profile picture now has three honest states in Channels → Widget → Style: upload your own, inherit your workspace logo, or remove it — in which case the chat header shows the Asks mark instead of forcing an avatar on you.
Themes got stricter, in a good way: per-color customizations now apply only when your theme is Custom. Light and Dark render their designed palettes exactly — no more dark composer leaking into a light widget — and colored surfaces pick their own contrasting ink automatically. The greeting field grew into a proper multi-line editor while we were in there.
Waiting at three bouncing dots is over. The agent's replies now arrive the way they're written — streaming in word by word, in the website widget, the hosted chat page, the dashboard's widget preview, and both playgrounds.
And when the agent needs a moment to check your catalog or policies, it says so: a status line — “Writing a reply…”, “Making sure everything's right…” — in the visitor's language. Answers that assert facts hold back until they're checked, so what streams is what stands.
Ask about a product and the answer now lookslike a product: cards with the image, title, price, and a button — not a wall of text with “View Product” pasted three times. Text bubbles can carry their own buttons, and quick-reply chips suggest the next step.
Every channel gets the version it does best: the widget and chat page render full cards, WhatsApp gets its native tappable button and a tight card set, Instagram and Messenger get a swipeable carousel, and text-first channels like email get clean inline links. The inbox and the widget preview render exactly what the customer saw.
We rebuilt the agent's engine room this month, and the visible result is simple: answers that used to take 20 to 36 seconds — the ones where the agent checks your catalog, your policies, your knowledge base — now typically land in 7 to 9. The final reply is composed while its facts are being double-checked instead of after, searches the agent will obviously need are started early, and clean results skip a whole round of deliberation.
Accuracy didn't pay for the speed: verification was recalibrated so the agent stops refusing to state things your store page plainly says, while deterministic guards on links and claims run on every single reply.
The build step of onboarding used to show three rows that all ticked over at the end. Now each row is a real phase reporting live: Reading your website, Learning your brand & catalog, Writing your AI agent — and the brand and persona tracks run in parallel, so the whole step finishes sooner.
It's smarter about what it reads, too: your homepage plus a handful of key pages chosen intelligently — not a blind crawl — and for Shopify stores the catalog and policies stay live from the store itself. Logo detection now catches inline SVG logos, the agent's suggested questions seed your widget's conversation starters automatically, and the build keeps moving even if you wander off to another tab.
Installing Asks from the Shopify App Store used to end at a signup form. Now it ends in your workspace. Click Connect to Asksinside Shopify admin and — if your Shopify email is verified — you're signed in directly: no second account, no password to invent, and Asks picks the right workspace on its own when the store is already connected.
Security didn't bend for the convenience: unverified emails still sign in classically once (then the link is permanent), and two-factor prompts appear inline instead of bouncing you to a login page. The embedded app home now live-updates from Connect to Connected to Live without a refresh.
Your support tool shouldn't need translating. The dashboard is now fully localized in Spanish and Brazilian Portuguese — every page, every API error message, every notification and billing email, several thousand strings deep. Pick your language in Settings and it follows your account across devices.
Emails to your customers are smarter than a global switch: a “we replied to your conversation” email goes out in the language of that conversation, not the language your dashboard happens to be in.
The little round button on your site is the most-seen pixel of your support experience, so it stopped being configurable into incoherence. The launcher now always derives from your primary color, with an automatically contrasting mark on top — or your own icon: upload a square PNG or SVG under Channels → Widget → Style and it renders over your brand color, falling back gracefully if the image ever breaks. The old separate launcher color and six-icon picker are gone, and so is the flash of default blue before your config loads.
Also gone: the pre-chat form. Nothing stands between a visitor and their first message anymore — the agent collects names and emails conversationally, when it's actually the right moment.
A new workspace used to be a tour of empty tables. Now every pristine surface — the inbox, customers, each knowledge base tab, the AI agent pages, analytics, channels — explains what it's for, offers one button that takes you to the thing that fills it, and links the matching docs page. Search and filter results keep neutral copy; nothing shouts “set up your widget!” at you for filtering too hard.
Every module header gained a Read more in docs ↗link while we were at it, so help is one click away from the page you're actually on.
Leaving shouldn't require a support ticket. There's now a Cancel subscription button on the subscription page — for card-billed and Shopify-billed workspaces alike — that schedules the cancellation for the end of your paid period. You keep everything until then, your data stays put after, and a banner with a one-click Resume undoes the whole thing if you change your mind.
Shopify has no native “cancel at period end”, so Asks emulates one honestly: your plan runs its full paid term and is canceled at the provider just before renewal — you can never be re-billed into a plan you left.
The summary card in the inbox got a quiet overhaul. Summaries now read the wholeconversation rather than a truncated window, are written in the customer's language, and carry an honest AI-generatedstamp with a timestamp — so nobody mistakes machine notes for a colleague's.
Your judgment outranks the model's: correct a sentiment by hand and re-analysis won't overwrite it until the customer actually says something new. Topics you remove stay removed, an angrysentiment level joined the scale, and regenerating a summary has a short cooldown so the button can't be nervously mashed.
Integrations used to share one page where the good parts sat below two upgrade cards. There's now a proper Integrations group in the sidebar: Shopify, API Keys, Webhooks, and the Agent API each on their own page, each with room to breathe.
And paywalls stopped being mysterious: a locked page now shows a blurred preview of the actual feature behind the upgrade prompt — you see what you'd get, not a padlock and a promise. Locked means locked, too: gated pages fetch nothing until the plan allows it.
“How many megabytes is my shipping policy?” was never a question anyone could answer. Knowledge base limits are now counted in articles: 300 on Starter, 1,000 on Business, 3,000 on Premium — where an article is one thing you can point at: a written article, an uploaded document, a crawled page.
The meter in the sidebar reads the same way, updating a page during a re-sync costs nothing — only genuinely new pages consume allowance — and if you need more room without a plan jump, +100 articles is a $19/month add-on.
Navigation used to be two sidebars pretending to be one — an icon rail plus a sub-menu that morphed under you. It's now a single persistent sidebar with expandable groups: Analytics, Knowledge Base, AI Agent, Channels, Integrations, Settings. The active group opens itself, your expansion state survives reloads, and collapsed mode turns groups into tidy flyouts.
Details worth noticing: the Channels group shows each channel's brand icon with a live status dot, the knowledge storage meter lives inside the Knowledge Base group, and the inbox's customer panel is now collapsible — with its state remembered — so small screens stop feeling crowded.
For teams that promise response times, the inbox now keeps score. Define SLA policies per priority — a first-response target, a next-response target, a resolution target — measured in your business hours, not wall-clock time. Every conversation then carries a live chip counting down, in the list, the header, and the customer rail.
The point is the warning, not the autopsy: conversations flag themselves at risk before the deadline (25% of the target remaining, by default), alert the assignee, and fire webhooks your own systems can react to. New inbox filters surface everything at risk or breached in two clicks. Available on Business and Premium.
The AI could always answer in any language — but the widget around it spoke English. Now the whole frame is localized: English, Spanish, Portuguese, French, German, Italian, Dutch, and Indonesian, over a hundred interface strings each, translated by hand with the right register per language.
You choose which languages are enabled and which is the default under Channels → Widget → Behavior; visitors get their browser language when it's on your list. One deliberate rule: your greeting, your card titles, your conversation starters are your words — Asks never machine-translates them.
The right-hand panel in the inbox was rebuilt around what agents actually look for. Sections collapse, reorder, and hide — your arrangement persists — contact details copy on hover, and past conversations peek their last three messages on hover before you commit to the jump.
The new AI Agent section is the handoff package: how the conversation stands (AI handling, needs human and why, resolved and how confidently), the summary, sentiment you can correct with a click, detected language, topics, and everything the agent collected. Shopify customers bring a read-only card of their recent orders alongside.
The subscription page now defaults to yearly — $31, $119, and $279 a month billed annually, a straight 20% off — with monthly one tab away and nobody moved off the interval they already chose.
Plan changes got honest math: upgrades apply immediately with the prorated charge spelled out before you confirm; downgrades wait until the period you've paid for ends, shown as a pending change you can cancel with one click. No surprise invoices in either direction.
Even teams that keep a human on every conversation deserve the agent's knowledge. The inbox composer now has Copilot: Suggest reply(⌘J on an empty composer) drafts an answer from the same knowledge base and retrieval as the AI agent; rewrite actions polish a draft you've started; Summarize conversation condenses a long thread before you wade in.
Copilot is strictly advisory — it never escalates, resolves, or saves anything on its own, it works even where the autonomous agent is switched off, and drafts don't consume AI credits. The draft is yours to edit; the send button is still yours to press.
Attachments stopped feeling bolted on. Photos and videos in the inbox and the widget now open in a full-screen lightbox — arrow keys, thumbnail strip, downloads, a real video player. Media-only messages collapse into tight picture bubbles with the timestamp overlaid, the way messaging apps do it.
Documents grew up too: PDFs show a first-page preview in the conversation and open in an in-app viewer with a page rail — no more download-to-read. Drag, drop, or paste files into the composer with a caption, at up to 16 MB apiece.
Status menus mixing “open, pending, resolved, closed” ask you to answer three questions at once. Conversations are now simply Open or Closed. Whether the AI resolved it — and how confidently — is recorded as the outcome of closing; whether a human is needed lives on assignment, where it always belonged. The inbox filter reads All, Open, Needs human, Closed, and badges only mark exceptions.
For API consumers: the conversation.resolved webhook is now conversation.closed and fires from every close path; the API accepts the old status as an alias so nothing you built breaks.
Your page finally gets a say in when the widget opens. The embed now exposes a tiny JavaScript API: AsksWidget.open() and AsksWidget.close()for wiring any “Chat with us” button, link, or banner to the chat — and AsksWidget.versiontells you exactly which build your storefront is serving when you're debugging.
document.querySelector("#chat-with-us").addEventListener("click", () => {window.AsksWidget.open();});
If you run Asks on Shopify, your agent just got situational awareness. Through our storefront pixel, it now knows what the visitor was doing before they opened the chat — the product they were looking at, what's in their cart, the pages they browsed.
Ask “does this come in a bigger size?” and the agent knows what thisis. No more “could you tell me which product you mean?” round-trips. You can review the same context in the inbox under Store activity on any conversation, and turn the whole thing off with one toggle if you prefer.
The old Insights page answered “how many conversations did I get?”. The new Analytics section answers the questions that actually change your week: what are customers asking about, how do they feel, where is the AI struggling, and what should you fix first.
Two new destinations. docs.asks.app is a complete rebuild of our documentation — setup guides for every channel, deep dives on the AI agent, and a full public API reference, all in the same design language as everything else we ship.
And because trust is earned in the open, status.asks.app now tracks the health of the platform and the upstream providers we depend on, with incident history you can subscribe to by RSS.
Every person who talks to your agent — on any channel — now has one clean profile. The customer page pulls together identity, the channels they've used, collected details, satisfaction history, and every conversation, with a side rail that keeps the important facts in view while you read.
The index got smarter too: live counts and trends up top, and cross-links everywhere — jump from a conversation to the customer and back without losing your place. Shopify customers are linked to their store identity automatically.
One page trying to do three jobs never felt right. Your agent's knowledge is now organized the way you think about it: Articles you write, Websites we crawl, and Documents you upload.
Each website now has its own detail view listing every scraped page — what was learned, when it was refreshed, and what changed. A storage meter shows exactly how much of your knowledge quota you're using, and refresh alerts tell you when a source needs attention instead of silently going stale.
Support conversations aren't text-only, and now Asks isn't either. Customers can attach photos, voice messages, and documents on the widget, WhatsApp, Instagram, Messenger, Telegram, Slack, and Discord — and your agent doesn't just receive them, it understands them.
Setup used to be a form. Now it's a reveal. Paste your website address and Asks crawls your most important pages, learns your products and policies, detects your brand colors and logo, and builds a first version of your agent — then puts you straight into a live playground to talk to it.
No website yet? There's a manual path too: describe your business, upload a few files, and you get the same result. Either way, the first thing you do in Asks is the thing you came for — a conversation with your own agent.
Adding a website used to be all-or-nothing. Now Asks maps the site first and shows you every page it found — you tick the ones that matter and leave out the blog archive from 2019. Locale duplicates are filtered automatically, and Shopify product catalogs are handled through the product feed instead of page-by-page crawling.
Under the hood, crawling moved to a much more capable engine, and re-syncs now compare content fingerprints — only pages that actually changed get re-learned, so refreshes are fast and your quota isn't wasted on unchanged pages.
Trust in an AI agent comes down to one question: where did that answer come from? Answers now cite their sources — the exact articles, pages, or documents the agent drew from — right in the conversation.
Behind it sits a rebuilt retrieval pipeline: semantic and keyword search run together, a reranking stage promotes the passages that actually answer the question, and an automated groundedness judge scores every answer against the knowledge it cited. The result in our evals: substantially more accurate answers, and a paper trail for every one of them.
When AI and humans share an inbox, the one thing that can't be ambiguous is whose turn it is. Escalations and AI resolutions now show up as clear markers inside the conversation, every human reply carries the sender's name, and a Needs human badge — with a matching filter — puts the conversations waiting on your team front and center.
The customer panel on the right was rebuilt too: sentiment, CSAT, an AI summary of the conversation, and any details the agent collected — the things you used to scroll for, now at a glance.
Not every conversation starts on your website. Every workspace now has a hosted Chat Page — your agent on a clean, full-screen page at your own link, wearing your brand. Drop it in an Instagram bio, an email signature, a QR code on packaging — anywhere a link fits, your agent now fits.
It's the same agent, same knowledge, same inbox behind it. Find yours under Channels → Chat Page.
The oldest support channel finally gets the AI treatment. Forward your support address (say, help@yourstore.com) to your Asks inbound address, verify your domain, and your agent starts answering email — with proper threading, so replies land in the same conversation every time.
Emails flow into the same inbox as every other channel, escalate to your team the same way, and outbound replies are sent from your own domain once verified.
Your customers don't all live in one app, so your agent shouldn't either. Three channels joined the lineup:
Together with the web widget, WhatsApp, Instagram, and Messenger, that's one agent answering everywhere — and one inbox where your team sees all of it.
Your Asks agent knows your product better than any document does — so we made it callable from outside Asks. Every workspace can now expose its agent as a remote MCP server: connect it to Claude, Cursor, or any MCP-compatible client with an API key, and ask it questions from wherever you already work.
Your engineer can ask the agent about refund policy edge cases from inside their editor. Your ops team can query it from Claude while writing macros. Same knowledge, same guardrails — new front doors. And it works the other direction too: your agent can call your MCP servers as tools, configured under AI Agent → MCP Servers.
{ "mcpServers": {"acme-support": {"url": "https://api.asks.app/v1/mcp","headers": { "Authorization": "Bearer ask_..." }}} }
Everything the dashboard can see, your code can now see too. The public REST API covers conversations, customers, tickets, tags, channels, and agent configuration — authenticated with scoped API keys you manage in Settings, so an analytics script can't rewrite your prompts.
Prefer push over poll? Register a webhook endpoint and Asks sends signed events the moment they happen — new conversation, escalation, resolution, lead captured. Delivery logs, test events, and one-click secret rolling included.
curl https://api.asks.app/v1/conversations \-H "Authorization: Bearer ask_live_..."{ "data": [ { "id": "cnv_8f2...", "channel": "whatsapp","status": "resolved", "resolved_by": "ai" } ] }
Knowledge answers questions; actions solve problems. Custom Actions let you describe any HTTP API to your agent — an endpoint, its parameters, when to use it — and the agent calls it live, mid-conversation, when a customer needs it.
“Where's my order?” becomes a real lookup against your system, not a canned deflection. Every action runs with guardrails: response filtering so the agent only sees the fields you allow, secret redaction, rate limits, and an egress guard that keeps requests pointed only where you aimed them.
Nobody likes a form popping up mid-conversation. Turn on Data Collection and your agent gathers what you need — name, email, company, any custom field — the way a person would: by asking at the right moment and remembering the answer.
Details are saved to the customer profile as the conversation happens, visible in the inbox side panel, and exportable to CSV. Set which fields matter and the agent handles the rest.
An escalation at 3 a.m. shouldn't sound like one at 3 p.m. Set your business hours — timezone-aware, per weekday — and the agent adjusts how it hands off. During hours: “looping in the team now.” After hours: an honest heads-up with your actual next opening time, so nobody sits refreshing a chat that won't answer until morning.
Everything lives in the new Escalation hub under AI Agent, alongside your escalation triggers and handoff notifications.
Your widget is a guest on your website — it should dress for the occasion. Theming was rebuilt around three deliberate choices: Light, Dark, and Custom, where every surface, bubble, and accent is yours to set. Accent colors are applied where they belong — on actions — so the result reads as designed, not decorated.
The settings page shows a live, pixel-accurate preview as you edit, and the launcher is configurable too: size, label, position.
For teams with an identity provider, Asks now speaks SAML: SP-initiated single sign-on, just-in-time provisioning gated to your domain, an enforcement mode that makes SSO the only door in, and a break-glass owner path so a misconfigured IdP can't lock you out. Certificate-expiry alerts and an SSO audit log round it out.
For everyone: two-factor authentication on your account, a sessions page that shows every active device with one-click revocation, and email alerts on new-device logins and repeated failed attempts.
Some customers want to chat; others want to read. The new Help Center publishes your knowledge as a public, searchable site — organized into collections, styled to your brand, hosted by Asks at your own help URL.
It's one knowledge base wearing two hats: the same articles that power your agent's answers are the ones customers can browse themselves. Write once, help twice.