Conversations now live at their own URL. Refreshing keeps you in the same chat, Back and Forward behave as you would expect, and you can bookmark or share a link that opens exactly the conversation you meant.

Switch to another chat while Zoe is working and come back — she is still going, live, right where she was. Her thinking, tool steps and canvas carry forward instead of freezing at the last saved moment. Open the same conversation in a second tab and you will watch it stream there too.

Wi-Fi drops no longer end the conversation. Zoe shows "No internet. Reconnecting" instead of a bare error, keeps working while you are offline, and picks the answer back up the moment you are reconnected — no retyping, no lost work.

Coming back to a backgrounded tab on your phone no longer reports a network error that never happened.

Pressing stop now leaves a receipt: "Stopped by you after 6 steps" rather than a blank space. The research Zoe already finished stays on screen instead of vanishing along with the part you cancelled.

Reopening a conversation used to quietly lose things: tool results came back empty, canvas reports vanished, and Zoe's private thinking sometimes appeared as her answer. History is now drawn by the same engine as the live stream, so a reopened chat is identical to the one you watched — every thinking step, tool card, chart and image in place, in order.

When Zoe needs your approval she now says "Waiting for you" instead of showing a spinner as though she were still working. The chat title carries an amber dot so you can spot it from the sidebar, and it stays put when you switch away and return.

Coming back to a running task no longer restarts its clock at zero. A task that has been running four minutes says four minutes, counted from the moment it actually began.

Rebuilt around one thing: what you want to do next. A short greeting, the composer, and five suggestions — Research, Trends, Competitors, Strategize, Create. Everything else moved into Explore, which now slides up as a panel over your conversation instead of taking over the screen.

Zoe greets you by name and by the hour — morning, afternoon, evening or late night — and mixes in a lighter word when you are mid-flow. It updates itself if you leave Zoe open.

New chat now looks like the action it is, and the menu follows how you actually work: Canvas Gallery, Scheduled Tasks, App Store, API Keys, Zoe Credits, What's New.

Canvas reports no longer appear as raw code in the chat. Tool cards no longer spin forever when Zoe runs several at once. Suggestion and composer icons share one consistent style. On phones and with the canvas open, suggestions sit under the greeting and the input stays at the bottom where your thumb is.

Scheduled Tasks got a complete makeover. A brand-new home shows every timer, reminder, and recurring task in one place — create one with a simple form, or let Zoe set it up with you in chat. Need inspiration? Start from ready-made ideas like Weekly performance report, Competitor watch, or Daily content idea.

And schedules finally speak your language: "every weekday at 8 AM, until Friday" — Zoe stops exactly when you told her to.

Every scheduled task now keeps a run history. Click any run and Zoe takes you straight to that conversation, right at the result — no more hunting through chats to find what happened. Pause, resume, or run any task instantly from its page.

We rebuilt the engine behind timers and scheduled tasks from the ground up. Reminders now fire exactly once — no more repeated messages — arrive on time, and stream live into your chat along with a notification. Set multiple reminders? They all run side-by-side without waiting on each other.

Scheduled tasks can now generate content on their own — ask for "3 fresh post images every morning" and they'll be waiting for you. And your safety net stays absolute: anything that publishes to your real accounts always waits for your approval. Zoe does the work; you stay in control.

Tapping Approve is now bulletproof. We rebuilt the approval system end-to-end: no more "not approved" right after you approved, no more cards stuck on "Publishing…", no more wrong step counts. Your approval lands every time — whether you respond instantly or come back an hour later.

Approval cards are also clearer now: image and video requests say exactly what they'll do — "Generate image", "Generate video" — instead of a confusing "Ready to Publish?" for everything.

Dozens of reliability fixes — generated images always show up in your chat, saving to the Canvas Gallery is bulletproof, app connections fail gracefully with clear messages, and long research sessions with big data from your connected tools no longer stumble. And those rare moments where Zoe went quiet without answering? Gone — she now recovers on her own or tells you plainly what happened.

You asked for design tools, CRM, email marketing, WhatsApp, SEO, ad platforms, website builders — Zoe now connects to all of them. Over 100 production integrations are live in Zoe's App Store across design, analytics, advertising, communication, CRM, social media, and more.

Connect tools like HubSpot, Mailchimp, Semrush, Notion, Slack, Salesforce, Shopify — and Zoe can use them directly in your conversations. Create a design, send a campaign, check analytics, update your CRM — all from one place.

This is what makes Zoe your complete Marketing OS — not just a content tool, but the brain that connects and runs your entire marketing stack.

Zoe now runs on Google's latest Gemini 3.6 Flash — the most powerful and fastest model yet. You'll notice faster responses, sharper writing, better understanding of complex requests, and more accurate results across everything Zoe does.

Ever hit a wall mid-conversation where Zoe stopped remembering what you discussed? That's gone. We've made deep optimizations so Zoe can now handle unlimited back-and-forth in a single session — no matter how long your conversation gets. Plan a full campaign, refine it 20 times, and Zoe stays sharp from start to finish.

Zoe's App Store now lets you connect to any external tool in seconds. Just paste a URL, and Zoe automatically detects the right settings. Works with Cloudflare, Notion, Slack, GitHub, and hundreds of other services — no setup guides needed.

Tools that require login (like Cloudflare, HubSpot, or Salesforce) now work seamlessly. Zoe detects when a tool needs your permission, opens a secure login window, and connects automatically — just like linking a social account. Your credentials are encrypted and never stored in plain text.

When Zoe creates long documents, blogs, or reports, the canvas now smoothly scrolls along as content appears — so you can watch it build in real-time. Scroll away to read, and it respects your position without jumping back.

When Zoe edits a specific section of your canvas (like fixing a paragraph or updating a heading), the canvas now automatically scrolls to the exact change — no more hunting for what was modified.

Hero images at the top of blogs and documents now render as full-width, edge-to-edge banners — no more tiny centered images with empty space. Your content looks polished from the first glance.

Opening past conversations no longer shows "Accept/Reject" prompts for changes you already approved. History loads cleanly — exactly as you left it.

Fixed an issue where canvas components could duplicate during edits — especially with long documents or rapid changes. Your content now stays exactly as intended, every time.

Zoe's 45+ marketing tools are now available as an MCP (Model Context Protocol) server. Connect your favorite AI assistant — Claude, Antigravity, Claude Desktop, or any MCP-compatible client — and access analytics, posting, inbox, image generation, and more directly from your workflow.

Generate and manage API keys right from the sidebar. Each key has customizable scopes (analytics, posting, generation, inbox), expiry dates, and one-click revocation. Copy-paste ready config snippets for Claude, Antigravity, and other tools.

Your MCP connection URL is mcp.zocialone.ai/mcp — a clean, branded endpoint with SSL. Works with both header-based auth (desktop clients) and URL-based auth (web connectors like Claude). All 7 platforms, one endpoint.

You can now connect your social media accounts directly through Zoe's chat. Just say "link my Instagram" and Zoe opens a secure OAuth window. Supports Instagram, YouTube, Threads, LinkedIn, Pinterest, Facebook, and Google Business — all with official platform login flows.

After authenticating, Zoe shows you all available pages and profiles in a clean selection card. Pick exactly which accounts to link, skip ones already connected, and see instant success confirmation — all without leaving the chat.

Zoe now uses interactive input cards to collect your preferences before creating content. Pick platforms, choose tone, set schedules — all through clickable cards right in the chat. No more guesswork.

Before posting or generating any image or video, Zoe now shows you a preview and asks for your approval. Nothing goes live without your explicit OK. Review every result before it's finalized.

Click any element on a canvas to leave a comment — just like a design tool. Tell Zoe exactly what to change on a specific section, and it'll update that component directly. No more describing which part you mean.

The new Zoe App Store lets you connect custom MCP tools and third-party integrations. Extend Zoe's capabilities with your own APIs, databases, or specialized tools — all plugged in through a simple interface.

Set one-time timers or recurring schedules and Zoe will run tasks automatically in the background — daily reports, weekly analytics summaries, content reminders, and more. She works even when you're not in the chat.

The all-new Omni video model can take your audio, images, or text and generate stunning 10-second videos with custom visuals and motion. Perfect for Reels, Shorts, and Stories.

New Nano Banana Lite model generates images 3x faster with great quality for quick drafts and social posts. Zoe picks the right model automatically based on your needs.

When Zoe updates your canvas, every change is visually highlighted — green for new, yellow for modified. The review bar shows a component-level breakdown so you can accept or reject with confidence.

Zoe now plays a subtle audio chime when she finishes processing your request, needs your approval, or is waiting for input. Switch tabs freely — you'll always know when she's ready.

New tools for LinkedIn competitor ad tracking and Meta ad library search. See what ads your competitors are running, analyze their messaging, and get actionable insights to outperform them.

Find potential brand partnership opportunities across Meta and LinkedIn. Zoe can now identify complementary brands, analyze their audience overlap, and suggest collaboration strategies.

Clicking links in Zoe's responses now opens in a new tab instead of navigating away. Your conversation stays exactly where you left it.

Fixed full-screen image preview on iOS — the close button is now always visible, even on iPhones with the Dynamic Island.

Click any element in the canvas to add inline comments — just like you would on a design tool. Write your feedback, click Submit, and Zoe will read all your comments and update the canvas accordingly. Perfect for iterating on dashboards, strategies, and content previews.

When Zoe updates your canvas, you'll see a review bar showing exactly what changed — which components were added, modified, or removed. Click Accept All to keep the changes, or Reject All to instantly undo them. Changed components get a visual highlight so you can spot differences at a glance.

The Canvas Gallery now shows real screenshots of your canvases instead of placeholder icons. Every canvas version is saved with a timestamp so you can browse your history and restore any previous state. Dates now display correctly as relative time (e.g., "5m ago").

Canvases now build component by component as Zoe generates them — you see the dashboard forming in real-time instead of waiting for the entire response. Combined with smarter patch detection, follow-up edits are instant instead of rebuilding the whole canvas.

Fixed an issue where shared canvas links opened to a blank page. Shared links now load correctly with full styling and interactivity — anyone with the link can view your canvas.

Zoe now uses Google's Interactions API — a next-gen streaming engine that makes conversations faster and more reliable. Multi-turn chats use less bandwidth, tool calls are more efficient, and you'll notice smoother streaming overall.

When Zoe is reasoning about your request, you can now see her thinking process in real-time. A collapsible "Thinking" panel shows her analysis — great for understanding why she made a creative decision or chose a strategy.

Zoe can now create cinematic AI videos from text using Veo 3.1. Describe your brand story and get a stunning 10-second video — perfect for Reels, Shorts, and TikTok. Full creative direction with camera movements, lighting, and mood control.

A completely redesigned creative knowledge base with 12 reference guides covering prompt engineering, image templates, video production, and industry-specific styles. Zoe's image and video output is now dramatically more professional and on-brand.

Every single step of Zoe's workflow — from model calls to tool execution to content delivery — is now fully instrumented with error tracking. If anything goes wrong, our team sees it instantly and can fix it before you even notice.

Zoe now has her own App Store. Browse 30+ pre-built integrations — Google Calendar, Notion, Slack, Zoho CRM, HubSpot, Stripe, and more — and connect them with one click. Zoe can then use those tools directly in chat: pull CRM data, check your calendar, send Slack messages, all through natural conversation.

Click any connected app to see every tool it provides — with descriptions explaining what each one does. Toggle individual tools on or off, use Enable All / Disable All for quick control, or remove the entire app connection. You decide exactly what Zoe has access to.

Built your own tools? Connect any custom MCP server by providing a URL — Zoe will auto-discover all available tools and make them available in chat. Supports Streamable HTTP and SSE protocols with optional bearer token or API key authentication.

Zoe now asks for your explicit approval before publishing any post, replying to comments, or taking important actions. You'll see a beautiful interactive card where you can review, edit, approve, or reject — nothing happens until you say so.

When Zoe needs your input — like picking a target audience, choosing a content tone, or selecting platforms — she now shows interactive forms with radio buttons, checkboxes, and text fields. Pick your answers with a tap instead of typing them out. You can even write in your own custom option.

Before any post goes live, you see a full preview card showing the caption, platforms, and hashtags. You can edit the caption right there in the card, tweak anything, and then approve. No more going back and forth.

If Zoe's connection drops while she's working on something for you, she now picks up exactly where she left off — including any pending approvals. No lost work, no repeated requests.

Zoe has learned specialized playbooks for every type of task — content creation, publishing, analytics, competitor research, inbox management, and more. She now instantly knows the best approach for your request without guessing, giving you faster and more accurate responses.

Zoe now only loads what's needed for each request. A simple message like "hi" or "show my analytics" that previously used 10+ credits now costs just 1-2 credits. Your plan stretches significantly further on everyday conversations.

New Thinking Depth selector next to the send button. Choose how deeply Zoe reasons about your request:

🟢 Minimal — Fastest, cheapest. Great for quick questions.
🔵 Low — Light reasoning, good balance of speed and quality.
🟣 Medium — Balanced reasoning (default).
🔴 High — Deep strategic analysis for complex tasks.

Lower depth = fewer credits used. Pick Minimal for simple chats, High only when you need deep strategy work.

Fixed an issue where some users were incorrectly shown "Quota Exhausted" even though they had credits remaining. If you have credits, you can chat. No more false blocks.

We've completely rebuilt Zoe's engine from the ground up. Everything you do — chatting, creating content, analyzing competitors, pulling analytics — is now significantly faster with fewer interruptions. Think of it as Zoe getting a brand new sports car engine under the hood.

Every conversation with Zoe is now fully authenticated and encrypted. Your marketing data, social media accounts, and business insights are protected with enterprise-grade security. Only you can access your Zoe workspace.

All your chats, canvas creations, and analytics are now permanently saved — even if Zoe gets updated or restarted. Pick up exactly where you left off, every time. No more disappearing conversation history.

We've built an automated quality pipeline that tests every improvement before it reaches you. This means new features arrive faster and bugs get squashed quicker — without compromising stability.

The entire Zoe interface has been optimized for speed. Pages load faster, switching between chat, canvas, and gallery feels instant, and the overall experience is noticeably smoother — especially on mobile.

The Ultra tier now includes Claude Opus 4.8 alongside Opus 4.6 — the latest in Anthropic's deep reasoning family. Enterprise users get both Opus models in a shared Claude pool for maximum flexibility.

The model selector offers 3 tiers: ⚡ Fast (Gemini 3.6 Flash), ✦ Pro (Gemini 3.1 Pro), and 💎 Ultra (Claude Opus 4.6/4.8). Each tier shows the underlying model name and a clear use-case description. Enterprise users default to Ultra.

Use Zoe on your phone or tablet with the same full experience as desktop — browse canvases, tap images to view full size, download media, and navigate seamlessly. Everything adapts perfectly to your screen.

When Zoe searches the web for you, results now appear as clean citation cards with the website's favicon, title, and a brief snippet — making it easy to see exactly where information comes from. Click any citation to open the source directly.

Every Zoe response now shows how long she took to work on it — "Worked for 14s" appears at the top of each reply. Click the arrow to expand and see the full breakdown of thinking steps, tool calls, and research she performed behind the scenes.

Web and news searches are now prioritised for India, so you get more relevant local results — Indian news sources, regional business insights, and market data that matters to your audience.

Zoe now gets back to you in record time! We've supercharged the underlying engine, slashing response times so you get instant progress and feedback on your marketing tasks without the wait.

Zoe now runs on Google's latest AI model, launched just yesterday at Google I/O 2026. Everything you do — competitor research, content creation, analytics — is now noticeably faster with sharper, more nuanced insights. Zoe can now handle longer conversations, deeper audits, and more complex requests without slowing down.

When you ask a follow-up question, Zoe no longer starts her analysis from scratch. She builds on what she already reasoned through — so your third question in a conversation gets an answer that's informed by everything discussed before. Strategies get sharper the longer you talk.

Fixed the recurring issue where competitor post images and videos showed "Media unavailable" placeholders in Canvas reports. Media is now validated before saving — broken or expired links are caught immediately instead of being stored as corrupt files. Your shared Canvas reports stay clean.

Profile pictures now display as compact circular avatars instead of stretching full-width. Logos and thumbnails in Canvas are sized naturally — no more oversized images taking up half the screen. Post links now correctly open the actual Instagram/Facebook post page.

Simplified model selection from 6 models to 3 clear tiers: ⚡ Fast (Gemini 3.0 Flash), ✦ Pro (Gemini 3.1 Pro), and 💎 Ultra (Claude Opus 4.6). Enterprise users now default to Ultra for the deepest AI reasoning. Each tier shows the actual model name alongside a clear use-case description.

Added a "Private Preview" badge in the sidebar to communicate early access exclusivity. Enterprise users see a dedicated badge on the welcome screen: "Powered by the industry's most advanced AI reasoning models." This is a handpicked preview — only select businesses have access.

Removed the logout button for embedded deployment. Hidden the auth debug dialog behind a developer flag. Sidebar now starts collapsed by default for a full-width chat experience. Voice options hidden by default with a toggle to enable. Streamlined for a distraction-free CMO experience.

Fixed a critical bug where voice chat messages disappeared from the screen the moment you ended a call. Voice sessions are now correctly saved to your chat history and remain visible after the call ends — same as text chats. Your voice conversations also appear in the sidebar so you can revisit them at any time.

Zoe now knows what time it is when you start a voice session. Your browser's local time and timezone are sent at connection, so Zoe greets you correctly — "Good morning" before noon, "Good afternoon" from 12–5 PM, and "Good evening" from 5 PM onwards. No more hearing "Good morning" at 1 PM.

Zoe's voice mode has been fundamentally upgraded from rule-based instructions to a reasoning-first architecture. Instead of following a fixed script, Zoe now reads the room in real time — adapting her tone, depth, and register based on professional seniority, industry context, and cultural cues. C-suite conversations are handled peer-to-peer; SMB founders get warmth and clarity. The upgrade also embeds deep marketing intelligence: neuroscience of social media engagement, platform-specific content mechanics, and India market context.

New stock media source: Pexels now joins Pixabay and Unsplash. Access millions of curated, royalty-free photos and videos with photographer attribution. New dedicated search_stock_vectors tool for SVG/vector assets via Pixabay. Multi-resolution URLs across all sources.

Every image and video in Canvas now has a download button (hover to reveal). Downloads use a server-side proxy to bypass cross-origin restrictions — works reliably in all browsers. Files save with descriptive names instead of UUIDs.

Download buttons now automatically upgrade stock CDN preview URLs to their full-resolution originals. Pexels images download at original quality, Unsplash at 2400px, and Pixabay at 1280px — not the small thumbnails shown in Canvas.

Stock video results now return all available resolutions (360p–4K) with FPS and file size metadata. Multi-resolution URL structure for quality selection. Download button on every video player.

Critical fix: scheduled posts now correctly convert from the user's local timezone to UTC. Previously, saying "schedule at 10 AM" could publish at the wrong time. Zoe now detects your timezone and converts automatically.

Smarter post status tracking: instant posts get mandatory status polling (the API save ≠ platform publish), while scheduled posts skip redundant polling. Clearer error reporting when platform publish fails.

Fixed dictation engine selector (Gemini vs Chirp) not persisting across sessions. Improved audio chunk handling for more reliable speech-to-text transcription.

Completely overhauled account-level and post-level analytics across all 7 platforms. Fixed critical data retrieval failures: Facebook page entity resolution, Instagram insights metric alignment, LinkedIn time-series chunking, YouTube quota-safe batching, Threads insights date handling, GBP performance metrics migration (v4.9 → Business Performance API), and Pinterest analytics pagination. All analytics tools now enforce 180-day query limits with automatic date clamping.

Post detail views now surface platform-specific engagement metrics: Instagram Reels (plays, skip_rate, avg_watch_time), Facebook videos (3s views, 1-min views), LinkedIn articles (click-through rate), YouTube (watch_time, avg_view_duration, card_clicks), and Threads (quotes, reposts). Each metric is correctly mapped to its platform API field — no more zeros or missing data.

Zoe now uses the fast platform-specific analytics tool (Instagram, Facebook, LinkedIn, etc.) when you ask about a single platform, and only falls back to the slower cross-platform dashboard tool for full overviews. Single-platform queries are now 10x faster.

Fixed the critical issue where canvas images expired after a few hours. All external URLs (Gemini CDN, Instagram/Facebook CDN, Unsplash/Pixabay) are now automatically downloaded and permanently stored when a canvas is saved. Competitor analysis screenshots and generated images now stay visible forever.

When you share a Canvas link on Teams, Slack, WhatsApp, or LinkedIn, it now shows a rich preview with the actual canvas title, description, and first image as thumbnail — instead of the generic "Canvas — Zoe AI" fallback. Powered by dynamic Open Graph meta tags.

Switching to List View in the Canvas Gallery now properly renders a horizontal card layout with the thumbnail on the left and title, description, and metadata on the right. Previously, the list view squished all content into a narrow column with missing text.

The "Copy Link" button in Canvas Gallery now generates proper shareable /canvas/ URLs with full OG metadata support. Previously used hash-based URLs (#canvas=...) that just opened the Zoe home screen instead of the canvas.

Fixed a major UI regression where LinkedIn's hashtag formatting logic was breaking chart rendering on the canvas. Disabled the aggressive text formatting on the frontend to ensure all charts, bar graphs, and UI components render flawlessly.

Two new tools: get_user_published_posts fetches your REAL published posts directly from platform APIs (all 6 platforms with analytics — Facebook, Instagram, LinkedIn, YouTube, Threads, Pinterest). Multi-page parallel fetching supports up to 50 posts. get_post_details drills into a single post for full engagement metrics. Ask Zoe "how are my Instagram posts doing?" to see it in action.

Zoe's mascot has been completely rebuilt with an animated octopus/jellyfish design featuring 12 expression states (sparkle, alert, wink, happy, sleep), blink transitions, organic tentacle movement, and ambient idle cycling. The mascot is now alive across the welcome screen, login flow, and thinking animation.

Video posts from competitors (Reels, Shorts, video ads) now render as proper playable video elements with play controls in the Canvas — instead of showing broken image icons. The system auto-detects video URLs from Instagram, Facebook, and YouTube CDNs and renders them correctly in both PostCard and Image components.

Fixed the "Media unavailable — CDN link expired" issue on shared Canvas reports. Instagram/Facebook CDN URLs contain signed hashes that get corrupted when passed through the AI model. Now, media URLs are stored server-side with short stable proxy keys (/api/media/cdn/...) — the AI never touches the raw URL, so competitor post images stay visible permanently in shared canvases.

Fixed get_post_details failing with 400 errors for Facebook posts. The Facebook API requires the full pageId_postId format, but Zoe was sending just the post portion. Now auto-prepends the page ID when missing — so "show me details on that post" just works.

Fixed search_stock_videos crashing with "Pixabay error: 400" when requesting 1-2 videos. Pixabay's Video API has an undocumented minimum of 3 results per page. Now automatically enforces this floor and trims results to your requested count.

AI-generated images and videos (from Nano Banana and Veo) now persist across server deployments. Previously, generated media was lost after each Cloud Run update because files lived on ephemeral container storage. Now all generated media is automatically backed up to Google Cloud Storage and restored on demand — so your Canvas posts, Reel covers, and generated videos stay visible permanently.

Fixed analyze_visual_content and analyze_video_content timing out when analyzing competitor posts via CDN-proxied URLs. The vision tools were resolving relative media URLs (/api/media/cdn/...) to the wrong server port in production, causing all visual analysis of Instagram/Facebook content to fail with "timed out" errors. Direct CDN URLs worked fine — now both paths work reliably.

We've integrated a powerful new suite of ADK FunctionTools hooked into Unsplash and Pixabay, bringing millions of royalty-free assets directly into the Zoe canvas. Here is how you can use them:

• Photos (Unsplash & Pixabay): Search real photography filtered by orientation and color palette. Use Cases: Build brand mood boards, find visual trend references (e.g., "flat lay photography"), or source background layers for custom design compositions.
• Illustrations & Vectors (Pixabay): Search for SVG icons, flat design characters, and scalable graphics with transparent backgrounds. Use Cases: Find consistent icon sets for carousels, grab infographic elements (arrows, charts), or discover seamless background patterns.
• Stock Videos (Pixabay): Pull high-quality 4K/1080p B-roll and atmospheric background footage directly into the canvas. Use Cases: Source B-roll for Reels/Shorts storyboards, find ambient background loops for Stories, or show examples of video styles to replicate.
• Editor's Choice: Toggle curated, high-quality selections for premium design inspiration across all media types.

We completely overhauled all 7 create_post tools with a "Zero-Trust" validation engine perfectly mirroring our production database and 2026 API constraints:

• Pre-Flight Validation: Model instructions now strictly validate character limits (e.g., Pinterest's 5,000 max description, Google Business's 1,500 offer terms, YouTube's 100 char title) BEFORE hitting the API.
• Parameter Instructions Updated: Detailed AI instructions for every input field. Added logical date validation for GBP events (Start < End Date, future only) and CTA URL validations.
• Native Tagging & Formatting: Enforced LinkedIn's Little Text Format (LTF) for hashtags ({hashtag|\#|Tag}) and added native dynamic @mentions for Instagram and Threads.
• Cron-Aligned Scheduling: Hardcoded a mandatory 15-minute future buffer for all scheduled posts, snapping times to 5-minute intervals (e.g., 5:05, 5:10) to perfectly align with ZocialOne's backend cron workers.
• Zero-Trust Confirmation: The agent is strictly blocked from publishing without presenting a pre-flight preview to the user for explicit confirmation.

Uploaded images and videos now display as a sleek media grid in your chat messages — with large thumbnails, video play overlays, and a "+N" badge when you attach more than 3 files. Click any thumbnail to open a full-screen lightbox with left/right navigation.

Adding or removing files before sending no longer causes the entire attachment bar to flash. Uses smart DOM-diffing — only the added/removed chip animates. File metadata (format, size) now shown for each attachment.

When using Claude, video analysis now works through the analyze_video_content tool — Claude has full agency to decide when and how to analyze. Videos uploaded once to Gemini Files API are reused (zero duplicate uploads). All vision tools now use unified Flash Lite model.

User-uploaded images now display correctly inside Canvas components (PostCards, Image blocks). Previously showed as broken icons when the model referenced local file paths. Videos in canvas also fixed.

Images and videos you uploaded now survive page refresh and session switching — media thumbnails are restored when loading chat history. Powered by a dedicated attachment store independent of ADK session state.

Switching between Gemini and Claude models now consistently persists across sessions. Resolved authentication expiration issues that caused the UI to silently revert to Gemini.

New dedicated analyze_video_content tool — Zoe can now deeply analyze any video (Reels, Shorts, ads) from URL or uploaded file. Scene-by-scene breakdown, hook assessment, branding analysis, and marketing effectiveness scoring. Works from any model including Claude.

Canvas Gallery search now properly filters cards visually (was showing count but not hiding non-matching cards). Also improved to word-based search — typing "vira reel" now matches "Viral Reel Concept" by matching each word independently.

Video uploads work seamlessly regardless of your selected AI model. Upload limit increased to 2GB via the Gemini Files API.

Videos attached to chat are now properly uploaded via the Gemini Files API for AI analysis. Previously, videos would silently fail and show only "Memory updated". Zoe can now watch, analyze, and discuss your video content across all Gemini models.

When a video can't be processed (too large, format issue, timeout), Zoe now tells you exactly what happened instead of failing silently. Videos up to 2GB are supported for analysis.

Deployed latest code with all recent improvements including Canvas Gallery, Dictation STT, Hey Zoe wake word, and Canvas History — ensuring production parity with development.

All AI-generated canvases are now automatically saved to a persistent, searchable library. Open it from the sidebar to browse, search, and revisit any strategy, analysis, or report Zoe has ever created — with visual miniature previews. Requested by Keerthi.

Search across the entire content of every canvas — not just titles, but every metric value, table cell, card label, and strategy detail. Find any insight instantly.

Each canvas in the gallery now has a "Copy Link" button — share direct links to any canvas with your team.

Click the mic, speak, and your voice is transcribed directly into the chat input. Two engines available: Gemini Flash for speed, and Chirp 2 for accuracy. Toggle between them with the STT selector.

Completely rebuilt wake word detection with fuzzy matching, audio chimes, voice readback, and a full command execution pipeline. Say "Hey Zoe, analyze my Instagram" — hands-free.

The canvas panel's "History" button now shows visual thumbnail previews of every canvas from the current chat — not just text labels.

Canvas components now render in real-time as they stream in! You no longer have to wait for the entire response to finish before seeing visuals. Overall tool performance has drastically improved.

We've eliminated the bottleneck of downloading media files locally. Image and video URLs are now passed directly to the UI and vision models, making visual analysis and display near-instant.

Canvas components now use a snappy, staggered fade-slide animation (0.25s) for a premium "building" experience. Auto-scroll to the bottom ensures ActionList buttons are always visible upon completion.

Fixed critical UTF-8 encoding issues that were crashing the thinking stream. The SSE pipeline now fully supports Claude 4.7 "adaptive" mode, allowing you to see its deep reasoning live during complex tasks.

Added a prominent "Chat with Zoe" CTA footer for shared canvases. Action chips on shared pages now automatically redirect visitors to the main app with the action pre-filled.

Meet the new Zoe — a cute iridescent jellyfish! The mascot now appears across the welcome screen, thinking animation, and login splash.

The Working timer now starts instantly on message send and persists until the full response completes. Timer freezes with a green "Worked for Xs" state — no more resets. Fixed a root cause where loadSessions was re-rendering the entire chat.

Refreshed Facebook and Instagram long-lived page tokens. All 6 platform tokens (Facebook, Instagram, LinkedIn, YouTube, Pinterest, Gemini) verified healthy for production.

Competitor posts now omit engagement metrics when the API returns no data, instead of showing misleading zeros. Native LinkedIn articles enriched via Open Graph scraping for titles and previews.

Fixed Unicode crashes across visual analysis, server logs, Instagram, and Places tools. All print statements now use ASCII-safe characters to prevent cp1252 encoding errors on Windows.

New amber ⚡ Working pill with live timer appears between the thinking panel and tool cards. Timer starts on first tool call and freezes to "⚡ Worked for Xs" when the response begins. Animated spinner turns green when done.

New get_linkedin_org_analytics tool — DAY, WEEK, and MONTH granularity time-series for ANY public LinkedIn page. Track impressions, reach, clicks, reactions, comments, and reposts for competitor trend analysis.

New get_linkedin_post_by_url tool — paste any LinkedIn post URL and Zoe resolves it to full post data (commentary, media, reactions, comments, reposts) via the API. Great for inspiration research.

Instagram Reels now report skip rate alongside plays and reach. Reposts count added across Instagram, Facebook, and LinkedIn for complete engagement measurement.

Fast visual analysis powered by Gemini Flash Lite — default context auto-fetches key attributes with a flexible focus_content field for targeted analysis. Instant competitor post breakdowns.

New community research tools — search Reddit, Quora, and niche forums for real audience conversations, pain points, and trending topics in any industry vertical.

Facebook, Instagram, and LinkedIn post tools now support since / until date filters. Pull posts from specific time windows — last 7 days, last month, or any custom range.

search_youtube now supports search_type: video, channel, playlist, or combos. YouTube posts return with inline comments. Channel profiles include unsubscribed trailers and thumbnails.

6 new tools: ad accounts, featured topics, trending products, product details, audience insights, and editorial articles. Carousel media now resolves properly. Deep visual commerce intelligence.

LinkedIn posts now properly resolve images, videos, documents, and carousel media. Instagram carousel items display correctly with individual media URLs.

Now correctly uses followers_count (people who Follow) instead of fan_count (people who Liked). Engagement rates use the accurate denominator.

Business details now include payment methods, accessibility features, amenities (Wi-Fi, parking, outdoor seating), and precise coordinates. Richer local competitor intelligence.

Zoe describes visual concepts and asks for approval before generating images or videos. No accidental credit consumption — you control every generation.

Updated system prompt with API accessibility matrix. Fixed tool progress widget and thinking panel rendering issues. Zoe now knows exactly which metrics are available vs restricted.

LinkedIn article detection, standardized metrics (Reach = uniqueImpressions, Reactions = likes, Reposts = shares). 78+ total tools across 7 platforms.

Fixed "Resource Exhausted" API errors, ensured clean server restarts, and resolved CSS visibility issues for the message copy button on the user side.

Zoe now reasons from 12,098 real YouTube comments, 3,787 videos, and 40+ academic studies — neuroscience of scrolling, virality formula (STEPPS), dark social, and emotional intelligence baked into every response.

3-tier visual generation: enhance real founder photos first → generate authentic-looking AI visuals second → surreal pattern-interrupts third. Always respects your creative vision.

3 new tools: audience comment analysis, video categories, and channel playlists. Now 69 total tools — all using 1-unit API calls for maximum data extraction within quota.

YouTube search now supports relevance_language for Hindi/regional content. Understands Jio effect, WhatsApp conversion funnels, and price-comparison virality patterns in India.

Replaced the top announcement banner with a subtle sidebar tag and slide-up changelog — less intrusive, complete release history in descending order.

Added one-click copy buttons for every AI response and tool call. You can now easily copy generated copy, strategies, or raw tool request/response JSON.

Added the ability to copy the entire Canvas text at once, plus a new "Share" button to generate public links so anyone can view your reports.

Fixed session timeout issues and connection drops during voice mode. Real-time Gemini Live conversations are now significantly more reliable.

Upload PDF, DOCX, TXT, CSV, and XLS/XLSX files directly to Zoe. PDFs sent natively to Gemini, text files parsed and injected, Word docs extracted via python-docx.

iPhone photos (HEIC format) now auto-convert to JPEG for processing. No more upload failures from iOS devices.

Generated images and competitor posts now appear in both the chat bubble and Canvas panel simultaneously. Zoe picks the best surface per query.

Web scraping: 15K → 50K chars. Tool output previews: 2K → 8K chars. Session history: 4x more content shown on switch.

Fixed crash when switching models (Gemini ↔ Claude) while a response is actively streaming. Now safely aborts the current stream first.

Real-world business prompts covering all 7 platforms — competitor research, content strategy, campaign planning, influencer outreach, and more.

Additional A2UI components for richer visual output. PostCard schema now supports embedded images. Chat freedom — no more forced 1-3 sentence limit.

Cloud Run instance memory doubled from 2GB to 4GB for handling larger file uploads, concurrent sessions, and heavier AI model processing.

Zoe CMO Agent with 9-point reasoning engine, 66 tools across 7 platforms, Canvas UI with 23 A2UI components, voice mode, memory bank, image generation (Nano Banana), and video generation (Veo 3.1).

Bidirectional voice conversations powered by Gemini Live API. 8 voice presets. Real-time competitor research during calls.

Long-term memory powered by Vertex AI. Zoe remembers your business, preferences, and past conversations across sessions.

Phone-based login with OTP verification. Linked social accounts with platform-specific data access.