Security update maintenance
Hello,
A maintenance operation is scheduled today from 12:30 p.m. to 2:00 p.m. to apply a security patch. The service may be temporarily unavailable during this time.
Thank you for your understanding.
Changelog
Follow new updates and improvements to Swiftask.
Two new models land on Swiftask: GPT-6 Luna and Claude Opus 5.5
Hello,
Your Swiftask workspace gets two new powerhouses at the end of this month, each with its own specialty. Here is how to choose in 30 seconds:
⚡ GPT-6 Luna
Fast, lightweight, and capable of processing up to 1 million words of context at once (roughly 1,500 pages).
Use it for:
High-volume processing: Sorting, categorizing, and tagging dozens of messages, tickets, or documents.
Summarizing and extracting key insights from long reports, contracts, or meetings.
Everyday queries: Information search, explanations, writing, and brainstorming.
Reading images: Screenshots, invoices, and table transcriptions.
🧠 Claude Opus 5.5
Anthropic's most powerful model to date: it reasons before acting, stays on track through multi-step workflows, and clearly explains its process.
Use it for:
Complex end-to-end projects that require execution across multiple steps.
Serious coding: Building features, debugging stubborn issues, or auditing entire projects.
Production-ready deliverables: Presentations, spreadsheets, and structured executive summaries.
You will also see Claude Opus 5.5 (Vertex AI) in your selector. Same model, same capabilities—routed through Google Cloud infrastructure, with the option for European hosting.
This is the recommended choice if you handle sensitive, personal, or regulated data (GDPR, healthcare, finance, public sector) where data sovereignty is just as critical as response quality.
How to get started?
No setup required: open the model selector in your Swiftask chat or under Agents—both models are live and ready to use. Enjoy exploring!
Accounts required: one single agent, each with their own account
Hello,
Here is a new feature in Swiftask that solves a problem many of you know all too well.
You create an agent that manages emails seamlessly, you share it with your 80 team members... and everyone ends up in your inbox. Until now, you had to choose: either keep the agent to yourself or recreate it 80 times.
The new “Accounts required” feature fixes this. Anyone who opens the agent logs in with their own account. A single agent, but working within everyone’s personal workspace.
🔐 Your data stays yours. Everyone only sees their own inbox and files.
📧 Every action carries the right name. An email is sent directly from the person who requested it.
🔑 Company permissions apply automatically. No extra setup needed.
🛠️ One single configuration, done once and for all.
Take, for example, a company using Microsoft 365 that connects an agent to Outlook to sort emails and draft responses. The administrator goes to Accounts required, shares the agent, and everyone logs in with their usual Microsoft account. That’s it.
And… it works with all external connectors in Swiftask: Gmail, Slack, Notion, Salesforce, HubSpot, and more.
The concept stays the same everywhere: a shared agent, individual accounts, and data that never overlaps.
📍 Where to find it: Agents > Configure Agent > Agent Instructions > Accounts required. Add the accounts to connect, and everyone will link theirs when they first use it.
Enjoy discovering this new feature on Swiftask!
Earlier updates
- Live view, Web navigation, and MCP: Your agents are leveling upA quick update on this week's new features 👋 We’ve added 3 key capabilities to bring more transparency and autonomy to your agents. 👁️ Live view and sub-agent memory: You can now see what sub-agents triggered by your main agent are doing in real time, instead of letting them run invisibly in the background. They also retain memory across sessions: resume your complex projects without having to explain everything again. (Enable in Agents ➔ Sub-agents) 🌐 Interactive Web Browser: Your agents are taking action directly on the Web. They open pages, click, fill out forms, and extract the exact data you need, completely on their own. Go to Agents > Configure Agent > Agent Instructions > Interactive Browser. Combine this capability with other skills for a complete, end-to-end workflow. Your agent can build a competitor price comparison table, send it by email, or save it straight to your Google Sheet. Add a trigger, and the task becomes a recurring automation. 🔌 Swiftask Code Connected via MCP: Developers, your VS Code extension now integrat
- Three new models available on Swiftask: GLM 5.3, GLM 5.3 Flash, and Muse Spark 1.3Hello 👋 Three new models are landing on Swiftask at the end of this week. 🤖 GLM 5.3 (Z.ai): The most powerful model for coding and complex reasoning. Reach for this one for long, technical tasks and large conversations. 🤖 GLM 5.3 Flash (Z.ai): The perfect companion for your everyday tasks. Ultra-cost-effective, fast, and multimodal—it analyzes your text, as well as your images and videos. 🤖 Muse Spark 1.3 (Meta): The efficiency specialist. It gets straight to the point (20% fewer steps) for faster responses. Plus, if your prompt lacks clarity, it asks you directly instead of guessing. All three models are already available in the Chat and in your Agents. Enjoy exploring them 😉
- New design and better organized menus for your settings — September 17, 2026 updateHi 👋 We've given a fresh new look to the Agent Configuration and Workspace Configuration areas, as well as the Agent Builder. The goal is simple: let you find your settings without having to search for them. ➡️ Reorganized menus: Your settings are now grouped by category across both agent and workspace settings. You can easily spot the option you need within a cleaner, more readable design. ➡️ Quick access with Ctrl + K: Even faster than the menus. Press Ctrl + K (or ⌘ + K on Mac) from your workspace to open a search bar that displays all available menus at a glance. We’ve also refreshed the search layout to make it clearer and easier to browse. ➡️ Redesigned skills journey screen: We’ve also overhauled this view. Browsing your agents' skills and adding a new one is now clearer and faster. ➡️ Clearer Maestro in the Agent Builder: When your prompt lacks detail, Maestro no longer guesses. Thanks to the Clarification skill, it asks for details directly within a redesigned interface and offers preset choices. You respond in one click, and it gets right back on track. No action needed on your end. Your settings remain th
- Knowledge tables that fill in automatically, views that adapt to you…Hi everyone, You’re likely familiar with this situation: your teams process dozens of requests every week, your AI agents perform exceptionally well, yet when it’s time to take stock, no one really knows where things stand. The data exists, but it’s scattered across chat histories, files, and spreadsheets that no one updates anymore. So you end up manually piecing numbers together or relying on gut feeling. That’s exactly what we wanted to solve this month, starting in the Knowledge section, specifically within your Knowledge tables. These are your business data points organized in rows and columns—leads, customer inquiries, applications, marketing campaigns—structured so your agents can read and use them to assist you. Until now, you had to fill them in manually. Not anymore. Your agents now write to them automatically as they work. For instance, an agent qualifying leads will update your sales pipeline while you focus on other priorities. Views adjust to what you want to see, and if you need a specific dashboard, all you have to do is ask. Your agents populate your tables on their own An agent can now write directly into a knowledge table. Instead of just reading your data, it enriches it continuously. Every processed request, analyzed document, or conversation automatically becomes a clean row in the right columns. This addresses a widespread issue: data generated during work often gets lost. Your agent would complete a task brilliantly, only for the details to sink into chat history. Counting, comparing, or identifying trends became nearly impossible. When making decisions, team members either guessed or spent days rebuilding numbers manually. This workflow applies to any continuous stream of incoming tasks—whether support tickets, marketing campaigns, or job applications. Take an HR department hiring for five open roles simultaneously. As the agent screens incoming resumes, it automatically populates a table with target roles, years of experience, key skills, and availability. No manual entry required. Three weeks later, instead of opening 120 resumes to make a shortlist, the hiring manager reviews an organized database and notices that top applicants are concentrated in only two roles. The other three listings aren't weak, they're under-promoted, an insight that passive reading would never have revealed. You only see what matters to you Your tables now support filters and groupings. Select your criteria, and the display reorganizes instantly. You can group rows by column—by owner, status, priority, and more—to digest information in structured sections rather than endless lists. This grouping functionality also unlocks Kanban views. Each group becomes a column, each row a card, and progress can be reviewed at a glance rather than decoded from a hidden status column. The same table can be viewed in two ways depending on your goal: a grid when you need to compare, count, and analyze; a Kanban board when you want to monitor workflow and identify bottlenecks. This eliminates scrolling through 800 rows to find the 12 relevant to you. More importantly, it removes the habit of exporting data to a spreadsheet “to sort in peace”, which inevitably created outdated duplicate files. Consider a customer support manager tracking 800 open tickets from the current quarter. Instead of reviewing everything, they filter by unresolved tickets and group them by issue type. Within seconds, a trend emerges: one-third of complaints stem from a single product line. Volume
- New chat widget release: smoother interactions, and more features for your visitors👋 Hello, and welcome back! Vacations are behind us, schedules are filling up, and your website is about to see traffic pick back up. Perfect timing: the chat bubble you embed on your web pages has just taken a massive leap forward. It is now built on the exact same foundation as the agent deployed within the main application. Anything you can do inside Swiftask, your visitors can now do directly in the chat bubble. Same features, same possibilities, same smooth experience. No more light version of the widget. What does this actually mean for you? This bubble is your Swiftask agent embedded directly on any webpage. Anyone who clicks on it needs no account, no setup, and no downloads: they open it, ask their question, and get an answer. Here are a few typical use cases: A Support Agent on your website: A customer runs into an issue at 10 PM on a Sunday. Instead of waiting until Monday morning, they open the bubble, describe their problem, attach a screenshot, and walk away with a solution. Your support team only handles what truly requires human attention. An HR Agent on your intranet: "How many days of paid leave do I have left?", "How do I report sick leave?", "Where can I find the remote work request form?". These questions pop up endlessly and pull your HR team away to handle low-value tasks. An agent connected to your internal documentation answers instantly, 24/7, without anyone having to open a support ticket. A Public Assistance Agent: For local governments, public administrations, or non-profits: the bubble answers citizens' questions regarding administrative procedures, opening hours, or required documents. Accessible to everyone, no account required, available around the clock. A Pre-sales Agent on your product pages: A prospective client hesitates and has three quick questions they would never bother submitting through a contact form. The bubble captures that exact moment. They get their answers, and you capture a lead that would have otherwise left without a trace. Real-time streaming responses This used to be a major pain point. Until now, visitors would send a message… and stare at a frozen screen. Five, eight, ten seconds of total silence, and then the entire response would drop at once. The issue? That visitor doesn't know you yet. Faced with a frozen screen, they don't think "It's processing"; they think "It's broken". So they either resend the message or close the tab altogether. Now, the response streams in right before their eyes, word for word, from the very first second. They immediately see that it's working, start reading right away, and waiting becomes invisible. See the Agent at work Your agents don't just generate text answers: they search the web, consult their knowledge base, query connected tools, or send emails. These actions take time, and until now, they happened entirely in the dark. Now, your visitors can follow the work in progress in real time. Whenever the agent triggers a skill, the action displays step-by-step as it happens. This turns waiting into an engaging experience: waiting ten seconds while seeing "Searching internal documents..." feels entirely different from waiting ten seconds in front of a blank screen. It also builds tr
- Enhanced security, better agent control, and new AI modelsBack-to-school on Swiftask! We've prepared some great new updates for you. A new season means new challenges and projects picking back up at full speed. To help you navigate this September with peace of mind, we’ve built some exciting improvements into your workspace: enhanced security, easier-to-customize assistants, and new AI models to explore. Here is a practical overview focused on what this changes for you day-to-day: 1. Access History for Shared Resources The sharing settings now feature a dedicated audit tab accessible everywhere across Swiftask—whether for an agent, a resource (document, folder), an artifact, and more. It automatically logs every access attempt and permission change. You maintain full visibility over the governance of your data and tools. At any time, regardless of the type of shared item, you can track permission changes (such as a role upgrade from Viewer to Editor, the exact timestamp, and who made the change) to keep your environment secure. Whenever you open the sharing options for an agent, document, or artifact, head over to the Access History tab. You'll find a complete chronological list of role modifications made by your team, ensuring every access right remains under control. 2. Simplified Custom Agent Duplication A new duplication interface allows you to clone an existing agent while selecting exactly which elements to keep. Save precious time when creating new assistants. No need to reconfigure everything manually—start from a proven baseline and adapt it in seconds. Use Case: To adapt a "France Customer Support" agent for the Spanish market, simply click Duplicate, preview its configuration, and swap out only its knowledge base with your translated documents. Your new assistant is ready to use immediately.
- Level up your power: 6 new AI models join your Swiftask workspaceHello, To give you even finer control over your projects and automations, we've just added 6 brand-new, cutting-edge language models to our catalog. Whether you're chatting directly or leveling up your custom AI Agents, these new LLMs bring faster speed, massive memory windows, and outstanding reasoning capabilities. Here is a quick guide to help you pick the right “brain” for your needs: ➡️ Claude Sonnet 5 (Google Cloud Platform) 🇪🇺: The coding and autonomous workflow specialist If you need to audit a complex codebase, fix bugs, or build automation scripts, this model automatically adjusts its reasoning level to carry out multi-step technical projects smoothly. It’s also a reassuring choice if you handle sensitive work data, as its deployment via Google Cloud EU infrastructure guarantees fully secure hosting and processing within Europe. ➡️ Scaleway DeepSeek V4 Flash 🇫🇷🇪🇺: Lightning speed on a sovereign cloud If you regularly deal with lengthy legal documents, project specifications, or hundreds-of-pages-long reports, this model leverages its massive memory window (1 million tokens) to generate summaries or retrieve specific information in record time. Since it is hosted directly in French data centers on the Scaleway sovereign cloud, you can process your most confidential content with complete peace of mind. ➡️ Qwen 3.8 Max: Massive memory for long-term workflows With an incredible processing capacity (up to 1 million tokens, equivalent to thousands of pages), this model is your ultimate partner for analyzing huge document archives or managing developer agents working over several days. ➡️ Mistral Large 3 & Medium 3.5: European excellence and visual analysis Mistral Large 3: Fluent in over 40 languages, it effortlessly analyzes your largest financial, legal, or technical reports in a flash. Mistral Medium 3.5: Versatile and agile, it handles both text and visuals (charts, diagrams, screenshots) while letting you adjust its reasoning depth depending on whether you need a quick answer or a deep dive. ➡️ xAI Grok 4.6: The specialist for long working sessions Designed to never lose track, Grok 4.6 excels at managing lengthy projects and creating visual interfaces. It’s ideal for building prototypes, refactoring code, or automating complex step-by-step financial analyses. How to start using them today? To chat directly: Go to the Chat and select your new model from the “Change AI” drop-down menu. To set up your assistants: Go to the Agents section ➔ Configure Agent ➔ Model Settings (LLM) to assign these new capabilities to your custom agents.