# Gemini Spark > **Status (October 2026):** renamed **Task** ("Task (formerly Spark)" in Google's help center). Needs Google AI Pro or Ultra. Not available in the EEA (Belgium included), Switzerland, the UK or Nigeria. Gemini Spark is [[Google]]'s personal AI agent, announced by [[Sundar Pichai]] at Google I/O on 19 May 2026 as "a 24/7 personal AI agent that helps you navigate your digital life". It turns the [[Gemini]] app from something you ask into something that works on your behalf: it runs on dedicated virtual machines in Google Cloud, so you "don't need to keep your laptop open". It was the first of the big-lab always-on agents. [[Grok Bot]] (August), [[Meta Muse]] (8 September), [[Microsoft Autopilot]] (25 September), OpenAI's [[Dots (AI)|Dots]] (29 September) and [[Claude Duet]] (21 October) all came later. ## The rename to Task In October 2026, Google quietly dropped the Spark brand. There was no blog post; the change rolled out server-side (on the web around 1 October per Chrome Unboxed, on Android around 12 October, on the Mac around 13 October). Spark is now a "Chat / Task" toggle next to the prompt box. "Gemini Spark Settings" became "Task settings", and a new "Scheduled" page lists your tasks (Active, Paused, Completed, Suggested). The plans page says "Tasks in Gemini". Google's reason, from its help page: beta users kept asking for help finishing "tasks". Chrome Unboxed and 9to5Google read it as Google de-emphasizing the Spark name. ## How it works Three building blocks: - **Task**: the goal you hand it - **Schedule**: when it runs, on a time or an event (daily news at 8 AM, an alert when your flight is delayed) - **Skill**: reusable instructions you call with @ or /. Skills are free and work in regular chat too The tools it uses: - A **remote browser** that keeps cookies and sign-ins between sessions, and that you can take over. It's only used on mobile - On desktop, your **local Chrome** instead, which has to stay awake (otherwise it may fall back to the remote browser) - A **remote computer** that runs code and keeps files (Markdown, code, data) between sessions - **Local folders** on the Mac, only the ones you explicitly link (and you can unlink them) A task thread sits next to a work panel that shows the planned, current and completed steps, plus the files it read or updated. You can run up to **15 tasks at once** (a schedule skips its run if 15 are already going). Usage limits are compute-based and shared with the rest of Gemini, higher on more expensive plans. The harness is [[Google Antigravity]], the same agent platform as Google's developer tooling. At launch, Google said Spark was "powered by Gemini 3.5" and called it "the first experience made possible by 3.5 models and Antigravity". [[Gemini 3.5 Flash]] was the only 3.5 model out that day, so it's the likely base, but Google never said so. Today you can change the model and thinking level per task; Google doesn't list which models are offered. ## What it does The I/O examples: parse your monthly credit-card statements for new or hidden subscription fees; read your child's school emails and send a daily deadline digest to you and your partner; turn meeting notes from email and chat into a Google Doc and a kickoff email. Josh Woodward (Google Labs) said small businesses use it to watch inboxes for customer questions. It doesn't only read and draft. It RSVPs and reschedules, replies to, forwards and labels emails, edits shared documents (with confirmation), builds spreadsheets and decks, manages Keep and Tasks, browses sites you're signed into, and runs code on its remote computer. ## Integrations What it connects to today: - **Google apps**: Calendar, Docs, Drive, Gmail, Keep, Sheets, Slides, Tasks, Contacts, Notebook, Google Photos (since September), Search, Finance, Flights, Hotels, Maps, YouTube - **Third-party apps**: Canva, Instacart, OpenTable, Dropbox, Zillow Rentals and more. The Canva, OpenTable and Instacart connections were announced at I/O, but Spark only started using them in early July - **Custom MCP servers**, now called "Connectors" ([[Model Context Protocol (MCP)]]) - **Chrome** (since 30 July, desktop, US first): auto browse with your logged-in accounts and saved passwords, with your permission Gemini as a whole lists 40+ connectable apps (13 October), most of them US-English only. ## Rollout | Date | What happened | |---|---| | 19 May | Announced at I/O; trusted testers. The Gemini macOS app ships to everyone the same day. Daily Brief launches separately | | 29 May | Beta for US AI Ultra subscribers, 18+ | | 30 June | Spark inside the Gemini Mac app (English, Ultra, 18+), with access to linked local folders | | Early July | Google Tasks, Keep, Canva, Dropbox, Instacart, OpenTable, Zillow Rentals; custom MCP servers; topic tracking (news, blogs, social, finance, sports, shopping, weather) | | 14 July | Ultra in all Gemini languages and countries, except the EEA, Nigeria, Switzerland and the UK | | Mid/late July | AI Pro in the US | | 30 July | Chrome integration (US first); AI Pro in 160+ more countries | | 3 September | Google Photos control (search, edit copies, private albums, share with approval), US Pro and Ultra | | 30 September | Skills in Gemini chat globally; Gems will be replaced by Skills (migration from 17 November) | | 1-13 October | Spark becomes Task | | 13 October | [[Gemini 4 Argon]] for Pro and Ultra; Antigravity built into the Gemini Mac app; a "Customize" hub with 40+ connected apps | ## Availability and pricing - **Requirements**: 18+, a personal Google Account (no work or school accounts "for now"), Google AI Pro or Ultra, Keep Activity turned on, and opted into Personal Intelligence - **Prices** (US): AI Pro $19.99 per month; AI Ultra $99.99 (5x Pro limits) or $199.99 (20x, down from $249.99 at I/O). The Ultra listing promises "first access to advanced features like Tasks in Gemini (in select countries)" - **Countries**: everywhere the Gemini app is supported, **except the EEA, Nigeria, Switzerland and the UK**. No timeline, no reason given - **Surfaces**: the Gemini mobile app, the Gemini Mac app and the web. Not the Windows app (launched 10 September) The European situation is odd. The Belgian Google page shows Pro at €21.99 and Ultra at €99.99 or €219.99, so EU subscribers pay full price and don't get Task. TNW pointed that out in September, and noted that an EC settlement requires comparable assistant access on Android by July 2027. ## Proactivity Task is proactive mostly through triggers you define: schedules, events, topic tracking and alerts. The Scheduled page also shows "Suggested" tasks. The morning digest is a different product. **Daily Brief** (Plus, Pro and Ultra, US, 18+) summarizes Gmail, Calendar and chats every morning ("Top of Mind", "Look Ahead"). Unlike [[Dots (AI)|Dots]] or [[Claude Duet]], Google doesn't describe Task as continuously scanning your apps on its own. That's my reading of the docs, not something Google states. ## Safety and privacy - **Approvals**: Task asks for review before sending messages, changing data, buying, submitting web forms or signing in with Google. It shows which sites it plans to visit, needs confirmation for shared-document edits, and hands passwords and payments back to you. One exception: bulk actions on your private Google Tasks happen without confirmation - **Google's own warnings**: prompt injection from websites, emails and documents; "safeguards don't guarantee protection". Google advises against sensitive tasks, recommends active supervision, and says Task may "make purchases or share your data with third parties unexpectedly" ([[Prompt injection]]) - **Training**: Task requires Keep Activity, under which Gemini activity is used "to provide, develop, and improve its services (including training generative AI models)", and a subset is reviewed by humans (kept up to 3 years). Google also collects "model thinking steps". The Privacy Hub doesn't say explicitly whether task and remote-computer data are used for training - **Third-party apps and custom MCP servers**: "Google does not monitor or secure" them - **Deleting data**: turning off task mode deletes the remote browser and remote computer data; tasks and schedules pause. Turning off Keep Activity does **not** delete that data **Memory**: there's no separate agent memory product. Task draws on Personal Intelligence (a mandatory opt-in), past chats, location, connected apps and signed-in websites; the remote browser keeps sessions, the remote computer keeps files, and Skills hold reusable instructions. ## Reception - **TechCrunch hands-on** (Sarah Perez, 30 May): "actually pretty useful". Good packing lists, deals and local activities. Faults: an invalid promo code, "five" read as "4-5", broken Google redirect links, price checks only every two weeks, and the separate Spark brand adding confusion - **TNW** (9 September): Europe excluded without explanation while paying the same prices - **Chrome Unboxed and 9to5Google**: the rename as Google stepping back from the Spark brand ## Why it matters Google's bet is that an agent reading your real data beats a cleverer model with no context. It starts with your Gmail, Calendar, Drive, Photos, Maps and Flights, plus Chrome with your saved passwords. It's also the cheapest way in among the big US labs ($19.99 for Pro, against $100+ for ChatGPT Pro with Dots). The same [[Google Antigravity]] harness that powers the developer tooling gets repackaged for everyone else. The weak spots: the least ambient proactivity of the group, a brand that changed after five months, and a whole continent left out. ## My take I'd like to test this one. The design (schedules you define, a work panel that shows each step, folders you link on purpose) is the kind of agent I can keep an eye on. I can't, though. Belgium is in the EEA, and the EEA has been excluded since day one, with no timeline and no explanation, while Belgian subscribers pay €21.99 or more for the same plans. And when it does arrive, I'd check the training terms first: Task requires Keep Activity, which feeds Google's model training. ## Caveats - Google never published a rename announcement. Dates come from Chrome Unboxed (web), 9to5Google (Android) and secondary sources (Mac). Parts of Google's help pages still say "Spark" - AI Pro in the US: Google's help-center changelog says 16 July, 9to5Google reported 23 July. The 14 July country expansion is known only through secondary quotes of that changelog, which now redirects to the release notes - Mac launch (30 June): Google says "supported countries", MacRumors says US-only. Spark itself was US-only then, so US is likely - Model: "Gemini 3.5" per Google. "3.5 Flash" comes from press. The current default model is unpublished - At I/O, Google promised emailing and texting Spark, custom sub-agents and local browser control for the summer. Local Chrome shipped; I found no primary evidence that email, text or sub-agents did. TechCrunch's launch piece says you can email Spark through a dedicated Gmail address, while Pichai said "soon" - "Runs 24/7" against the help page's "schedules won't run while task mode is off or your device is off": probably limited to local Chrome and Mac tasks, but the wording is ambiguous - The EUR prices come from a Belgian-localized page fetched by a tool; worth rechecking ## References - Google blog, "The next evolution of the Gemini app" (Josh Woodward, 19 May 2026): https://blog.google/innovation-and-ai/products/gemini-app/next-evolution-gemini-app/ - Google blog, Sundar Pichai's I/O 2026 keynote (19 May 2026): https://blog.google/innovation-and-ai/sundar-pichai-io-2026/ - Google blog, Google AI subscriptions (19 May 2026): https://blog.google/products-and-platforms/products/google-one/google-ai-subscriptions/ - Google blog, Gemini Spark July updates (30 July 2026): https://blog.google/innovation-and-ai/products/gemini-app/gemini-spark-updates-july-2026/ - Gemini release notes: https://gemini.google/release-notes/ - Gemini Help, "Task (formerly Spark)": https://support.google.com/gemini/answer/17094507 - Gemini Apps Privacy Hub: https://support.google.com/gemini/answer/13594961 - Gemini subscriptions page: https://gemini.google/subscriptions/ - TechCrunch, launch coverage (19 May 2026): https://techcrunch.com/2026/05/19/google-introduces-gemini-spark-a-24-7-agentic-assistant-with-gmail-integration/ - TechCrunch, hands-on (30 May 2026): https://techcrunch.com/2026/05/30/i-put-googles-24-7-ai-assistant-gemini-spark-to-work-and-its-actually-pretty-useful/ - 9to5Google, Ultra beta (29 May 2026): https://9to5google.com/2026/05/29/gemini-spark-ultra-us/ - 9to5Google, AI Pro (23 July 2026): https://9to5google.com/2026/07/23/gemini-spark-google-ai-pro-us/ - 9to5Google, Google Photos (3 September 2026): https://9to5google.com/2026/09/03/gemini-spark-google-photos/ - 9to5Google, Task redesign (12 October 2026): https://9to5google.com/2026/10/12/gemini-spark-task-redesign/ - MacRumors, Spark on Mac (1 July 2026): https://www.macrumors.com/2026/07/01/google-gemini-spark-comes-to-mac/ - Chrome Unboxed, dropping the Spark brand: https://chromeunboxed.com/google-is-quietly-ditching-the-spark-brand-to-merge-tasks-directly-into-gemini/ - Android Police, renamed Gemini features: https://www.androidpolice.com/googles-renamed-gemini-features-field-guide/ - TNW, Spark not in Europe (9 September 2026): https://thenextweb.com/news/gemini-spark-not-in-europe - PPC Land, EU and UK blocked: https://ppc.land/gemini-spark-blocks-eu-and-uk-users-as-google-adds-160-countries/ - Notebookcheck, saved Chrome passwords: https://www.notebookcheck.net/Google-Gemini-Spark-now-uses-your-saved-Chrome-passwords.1357683.0.html ## Related - [[Gemini]] - [[Gemini 3.5 Flash]] - [[Gemini 4 Argon]] - [[Google Antigravity]] - [[Google]] - [[Gemini Mobile App]] - [[Sundar Pichai]] - [[AI Agents]] - [[AI Agent Harness]] - [[Dots (AI)]] - [[Claude Duet]] - [[Meta Muse]] - [[Grok Bot]] - [[Microsoft Autopilot]] - [[Prompt injection]]