Changelog

See every agent and skill in your organization on one board, sorted by smart tags you describe in plain English that stay applied on their own, starting with Adoption and Efficiency.

Organization admins can now strip card numbers, Social Security numbers, passwords, and other sensitive details from everything an AI model sees.

Saving an agent or trigger now warns you when its instructions contain an API key, token, or other credential, and shows exactly where before you confirm.

Enterprise admins can now choose which websites an agent's browser can visit, or block specific ones.

Restrict an API key or token to the domains it's meant for, and agents can use it in requests to those domains without ever seeing the value.

Enterprise admins can now make a skill or knowledge source required on every agent, or a default that new agents start with.

Claude Fable 5.2, Gemini 4 Carbon, Gemini 4 Hydrogen, and Muse Ember are now available for your agents.

The Agents, Triggers, and Artifacts pages now show items from every team you are on, with a filter to pick teams and agents sorted by how often you use them.

Organization admins can add their own proxy, and agents can browse through it.

Turn on Channel Messages and the agent can answer every new message in its channels.

Deleting a shared agent or skill now shows everyone and everything it affects before you confirm.

Set Okta as your organization's agent registry and publish agents to its directory.

Turn on high contrast in Profile settings for stronger text, borders, and controls that improve accessibility.

The Gumloop app has been reorganized around collaboration across your agents and teams.

• Every team gets a Home with its stats, most used agents and skills, and connected accounts.
• Connectors, Skills, Brain, and Vault live together under Customize, for you and for each team.

Describe what a tool call must not do in plain language, and a decisions model judges each call against it to block, tag, or require approval.

Give an agent several accounts for the same connector, and it picks the right one for each task.

Make any earlier version of an artifact or skill the latest again.

Each agent's Performance page now shows which kinds of tasks cost the most and finish the least.

Add Gumball to a channel with /agents add gumball, and each person who mentions it gets their own Gumball with their own skills and connections.

Click a subagent to see its owner, setup, and the tasks it ran for this agent.

Ask an agent on the Evaluations home to find, explain, edit, or run any evaluation in your organization.

Scripts on a live artifact can now message, create, and list agents, as well as artifacts, skills, brain sources, and teams.

Agents on a call now run long work in the background and keep talking while it runs.

Words land in the message box as you speak, and holding Option + Shift + Space dictates without reaching for the mouse.

• Model versions are grouped into families, viewable by provider or by capability.
• Agent settings and the model picker show provider deprecation and retirement dates, and tell you which model actually runs when yours is retired or restricted.

Tag agents and skills in groups you define, filled automatically or by hand, and filter either table by them.

See who viewed your artifacts, how long they stayed, and what they did.

Describe what to check in a conversation and the Evaluation Builder sets up one evaluation that grades several agents.

Admins can move a member's agents, workflows, skills, triggers, and knowledge sources to a teammate in one go.

Agents are more reliable at signing in with logins you hand over and filling forms embedded from other sites, like Persona or Stripe, so they ask you to take over the browser less often.

Adding an existing skill to an agent now opens a preview, so you can read its files, edits, activity, and who uses it before you attach it.

Invite teammates to a shared trigger from its details page. Each person gets a Slack message or email, and accepting turns it on with their own accounts.

Artifacts on a custom domain now come with an embed snippet you can paste into any page. Artifacts can also use your microphone, camera, and clipboard.

Agents can now start a task when someone joins a Slack channel, like welcoming a new hire or sending onboarding docs.

Artifacts, skills, and agents can now be shared across teams.

Gumball can now set up custom MCPs for apps Gumloop doesn't have yet.

Both frontier models are available for agents today, Argon for long coding and research work and Wonder for long-running agent tasks and computer use.

Beyond writing and editing skills, agents can now set who can use one, rename it, or detach it from themselves, each on one approval card.

Reflections now reads an agent's prompt, skills, and triggers together and flags where two of them repeat or contradict each other, with a fix for each.

A custom trigger can now ask a decision model whether an event matters, like whether a message reads as an escalation, before it wakes the agent.

An agent can now send follow-up work to a subagent it already used instead of starting a new one, or hand work off and keep going without waiting.

Compaction now runs in the background, so an agent keeps answering while a long task is summarized instead of making you wait.

Text or Slack a voice memo to an agent and the transcript shows under your message in the chat.

Pick teammates from your Slack workspace, ranked by who you talk to most, and the invite lands in their Slack DMs.

Agents can now search your Wispr Flow notes, meeting notetaker, and calendar.

Add a secret to a custom agent once as agent-owned and everyone who can use the agent runs with it, including people outside its team on an org-shared agent.

Enterprise admins can now set a custom agent to always run incognito, so every task on it, in chat, Slack, or through the API, skips saving messages, tool calls, and files.

Ask an agent to rename a file, turn on its custom domain, or share it with someone, and it does all of it on one approval card. A file you have open refreshes when an agent changes it.

Video and audio files now play in the agent file panel and on the file page, with seeking, instead of a download-only preview.

Agents can now build out your Salesforce org, not just read it. Create objects and fields, run Apex, deploy metadata, assign permission sets, work approvals, and write records in bulk.

Agents can now look up events, leads, and transcripts in Sequel.

Anthropic's fastest Claude is now available for agents, with a 1M context and about a twentieth of Sonnet 5.5's price on prompts under 100k tokens.

Sharing, password link, and custom domain for an artifact now live in one Settings tab, and an artifact can keep team or organization access while also having a password link.

Agents can now look up accounts, contacts, and churn scores in ChurnZero.

Share a trigger you built and each person on the agent can turn on their own copy, running through their own account.

Scripts an agent runs on its Computer can now call AI models through the SDK, for per-row work like classifying every line of a CSV.

Enterprise admins can now build their organization's dedicated sign-in page from People & Access, with the URL, logo, brand color, and sign-in methods including SAML and Okta.

Agents can now manage courses, students, and enrollments in Skilljar.

Agents understand the conversation and know when a message is for them, even alongside other people and bots.

Agents now work with Amplitude, Bright Data, Buffer, Coda, Contentful, GoHighLevel, Higgsfield, Instantly, Navan, Neon, Responsive, Statsig, Tally, Vitally, Yotpo, and ZoomInfo.

Agent chats are smoother and much lighter on your laptop or phone.

Text Gumball over iMessage, RCS, or SMS to kick off a task from anywhere.

When an agent edits its own instructions, skills, or triggers during a chat, the changes are listed above the chat input and open with a click.

Messages you send while an agent is working are now steered or queued for you, so a correction lands right away and a follow-up waits until the agent finishes.

Sync a SharePoint site into Brain, with each document visible only to the people who can open it in SharePoint.

Agent tasks start sooner and long ones cost fewer credits.

Agents in Slack now use Slack's new agent blocks, so you see a task card with each step as it finishes.

Each agent trigger can now use its own model, so a weekly report task can run on Smartest while everyday tasks stay on Recommended.

Reflections now flag triggers that wake up and do nothing, routines worth turning into a skill, and tasks that ran far past normal, each with the credits it would save. Agents with auto-apply on apply confident edits on their own.

Agents can now read and send LinkedIn, WhatsApp, and email messages through Unipile.

Agents can now pull AppsFlyer attribution and campaign performance data for your apps.

Start a call with any agent or Gumball and just talk. Ask it to prep you for your next meeting, kick off a task while your hands are busy, or think through a problem out loud. Turn it on for your own agents in their settings.

Chats that finished or need your approval while you were away are marked unread until you open them.

Agents can now manage Campaign Manager 360 campaigns, creatives, ads, placements, and Floodlight tags.

Organization admins can open any member and see the connectors, models, and features they can use, and which custom role grants each one. New custom roles can start from no access, full access, or a template.

Agents now have two roles. Owners configure the agent and decide what Users can see and do, such as viewing team tasks, reading the instructions and connectors, or creating triggers. Users get a focused chat page and can request Owner access.

Send Now and Cmd+Enter on a queued message no longer restart the turn. The agent finishes its current step, reads your message, and keeps going.

Agents can now work with your Customer.io people, segments, campaigns, and messages across email, SMS, push, and in-app.

Agents can now pull LinkedIn Campaign Manager reporting: ad accounts, campaign groups, campaigns, creatives, and performance analytics.

Agents can now read and update leads, lists, programs, and campaigns in Adobe Marketo Engage.

Ask the Model Router which model to use for a prompt from your own code. Send the prompt and the models you have, and get back the pick and why.

Organization admins can register webhook endpoints under Settings, Data Export, Webhooks and get a signed request whenever someone joins the organization. Send a test event, watch the delivery log, and pause or resume an endpoint from the same tab.

Agents can now run and read Attest consumer research: brand tracking, creative and concept tests, and survey results.

Agents can now extract a design system from any site and check new designs against it with TasteLabs. Bring your own TasteLabs key.

OpenAI's newest image model is now the default for agent image generation. Agents can use your uploads or earlier images as references, and they pick the size, quality, and format that fit each request.

Organization admins can now set the starting model, reasoning, compaction, and access settings for every new agent from the Agents tab in organization settings.

When an agent finishes a reflection, its Slack DM now shows each suggestion as a card you can accept or dismiss right there. Notification links open the exact suggestion on the agent's Reflections page.

Agents can now search, read, and create your Evernote notes, notebooks, tags, and tasks.

Chatting with agents is much faster, from sending a message to seeing the reply.

Paste an artifact link into another chat, or hand it to a different agent, and ask for changes. The new version lands on the same artifact instead of becoming a copy.

Agent inboxes and Gumball now use shorter @gumloop.ai addresses. Mail sent to the old @gumloopagents.com addresses still arrives.

Turn on the Manage Evaluations ability and you can ask an agent to edit its own evaluation criteria or re-run evaluations on past chats. It checks with you before changing anything.

Open a Daily Chew as a chat and tell Gumball what it missed, and it remembers for next time. In Slack, hit Needs work on a brief to do the same in a thread. Works for Meeting Prep too.

Set reasoning effort and speed right in the model picker.

Restricted models in the picker now have a Request button, so you can ask for access without leaving the chat.

Select text in an agent's reply and quote it in your next message.

Drag agents in the sidebar into whatever order you want. Cmd+K follows the same order.

Agents can now pull calls, texts, contacts, and call transcripts from Quo, your business phone system.

Anthropic's most capable models are now available for your agents, and they're the strongest option for long-running work like deep research and complex coding. Organization admins can turn them on in model settings.

• Claude Fable 5.1

Google's newest Flash model is now available for your agents, with big gains in coding, agentic tasks, and multi-step reasoning at the same speed and price as 3.7 Flash.

Reply drafts can now look things up in your connected apps before writing, like checking your calendar before offering a meeting time or pulling what you discussed from your Granola meeting notes. You choose which connectors drafts can use from Smart Inbox settings.

You can now sync Gmail and Linear into your Company Brain. Emails, issues, documents, and projects get indexed, so agents can bring that context into their answers.

Agents can now generate images with Krea. Describe what you want in chat, enhance or upscale existing images, and iterate on styles without leaving the conversation.

Agents can now pull SEO data from Bing Webmaster Tools, like search performance, crawl stats, and keyword research, and submit URLs for indexing.

Agents can now work with Zip. Create and track procurement requests, check approvals, and look up vendors and spend from chat.

Agents can now generate audio with ElevenLabs. Turn text into natural-sounding speech mid-run, like narrating a report or drafting a voiceover.

Z.ai's newest model is now available for your agents. It approaches frontier models on coding and tool-heavy agent work at a fraction of their cost, and it understands images natively, so it can read screenshots, charts, and documents mid-task.

Organization admins can now serve shared artifacts from your company's own domain. Add it in organization settings, verify two DNS records, and dashboards, forms, and apps open on links like dashboard.yourcompany.com.

Agents can now browse the web in a real browser through Browserbase. Navigate pages, click buttons, fill forms, take screenshots, and pull data from sites without an API.

You can now sync Salesforce into your Company Brain. Pick which objects to include, like accounts, opportunities, and contacts, and agents pull that context into their answers.

Agents now complete long-running tasks much more reliably, and actions in your connected apps run faster.

Citations in Daily Chew and Meeting Prep briefs now point to the right source far more reliably. Click on any citation to see where an item came from and check the details for yourself.

Custom roles can now control who can create skills, and organization admins can automatically restrict newly released integrations, across the whole organization or per role.

Subscribe from Insights to get a recurring recap of what your organization or any agent you own accomplished. Reports cover tasks completed, actions in connected apps, top agents and users, and credit usage, delivered to your inbox or as a Slack canvas.

Agents can now run your Snapchat advertising. Create and manage campaigns, ad squads, and ads, upload creatives, build audience segments, track pixel conversions, and research public ads in the ads library.

Gumball can now find and use every skill shared with you or your organization, not just the ones you created. Mention a task and the right skill loads automatically, even if a teammate built it.

Agent pages now open noticeably faster. Chats appear right away without layout jumps, and switching between conversations no longer reloads the page.

Smart Inbox can now sort and draft replies for an Outlook mailbox, and Meeting Prep can brief you from an Outlook calendar. Pick Outlook when setting either one up.

Agents can now work with your Azure account. Review subscriptions and resources, manage compute, storage, databases, and web apps, check security and governance, and look up costs and pricing.

You can now switch an agent to Fast mode in its AI settings. Supported models like Claude Opus 5 respond much faster at a higher credit rate, and fast responses are labelled in chat.

Three new models to power your agents.

You can now sync Gong call recaps and transcripts into your Company Brain. Pick which workspaces to include, and everyone only sees calls they already have access to in Gong.

Enterprise organizations can now share agents with everyone in a connected Slack workspace, including teammates without a Gumloop account. Register the workspace in organization settings, then opt each agent in from its share settings.

You can now publish an artifact on its own URL, like sales-dashboard.gumloopartifacts.com. Turn on Custom Domain from the share dialog, pick a name, and hand out a link people can actually remember. Access and permissions stay exactly the same.

Agents can now report on your Meta advertising. Pull performance insights for campaigns, ad sets, and ads, review creatives, and read Facebook Page posts, comments, and reactions.

Adding a Company Brain source now shows an estimated credit cost before anything syncs. The sync only starts once you approve it.

Agents can now run SEO research through DataForSEO. Look up keyword volumes and difficulty, analyze competitors and backlinks, pull live search results, and track how often a brand appears in AI answers from ChatGPT and Google AI Overviews. Works out of the box, or connect your own DataForSEO account.

Agents can now work in Sprout Social. Pull profile and post analytics, monitor inbox messages and Listening topics, manage cases, and draft posts for publishing.

Agents can now work in Google NotebookLM Enterprise. Create notebooks, add and manage sources, and generate audio overviews and podcasts.

The largest open model released to date, US-hosted with a 1 million token context window and native visual understanding.

Agent chats now show a more accurate credit breakdown covering the conversation, tool calls, and subagents, so you can see exactly what used your credits.

Organization admins can now add a company Fireworks key, so open models like DeepSeek, Qwen, and Kimi run on your own account. Custom AI proxies are supported too.

Agents can now manage Reddit advertising through a new connector. Create and update campaigns, ad groups, and ads, build audiences and targeting, publish promoted posts, and pull performance and conversion reports.

You can now connect Snowflake with an RSA key pair instead of a password or browser login. This is the best fit for service users and accounts with MFA enforced, and it works everywhere Snowflake does in Gumloop, from agents to workflows.

You can now add and remove skills and connectors on your agents through the developer API. The same tools are available in the Gumloop connector, so an agent can even configure other agents.

Agents can now work in SharePoint through a new connector. Search sites and document libraries, manage files and folders, update lists, and publish pages.

We rebuilt the computer agents use to run Python, JavaScript, and shell code on new infrastructure, so code execution is more dependable, especially on long-running tasks.

SpaceXAI's newest model is now available in agents and workflows. It handles difficult, long-running tasks that require creatively using tools, whether in software engineering, data science, finance, or legal work, and solves multistep tasks in under half the steps of comparable frontier models.

You can now @mention a Brain source in an agent chat to point the agent at exactly the content you want it to answer from.

Brain sources on a team can now connect with a shared team account instead of someone's personal one, so the source keeps syncing no matter who set it up.

Organization admins can now see which connectors are in use across the company and who has connected each one, right from the Insights explorer.

Agents can now pull AppLovin campaign performance reports and manage your campaigns, creative sets, and ad assets through a new connector.

Skills now load onto your agent's computer up to 10x faster, so agents with large skill libraries get to work sooner.

Connectors on an agent can now use a credential owned by the agent itself, so everyone chatting with it runs on the same account instead of connecting their own.

AI steps in workflows are now billed by actual token usage instead of a flat rate, so short prompts cost less and you only pay for what you use.

The Outreach integration is now generally available. Agents can manage prospects and accounts, add prospects to sequences, and work with your emails, tasks, and calls.

Agents can now work with your Rocketlane projects and tasks through a new curated connector.

Proxied MCP connectors now support custom OAuth clients, so you can connect through your own app instead of Gumloop's.

You can now configure alerts for agent chat evaluations and get notified when conversations score below your bar.

You can now set a per-chat credit threshold on an agent. When a conversation crosses it, the agent pauses and asks for approval before spending more.

Agent owners can now see and manage the triggers teammates have set up on their agents, not just their own.

The DocuSign integration is now generally available. Agents can send documents for signature, track envelope status, create envelopes from your templates, and download signed documents.

Agents can now work with your ClickUp tasks, from creating and updating them to adding comments and organizing your spaces, folders, and lists.

Custom MCP servers now run more reliably after several under-the-hood fixes and improvements.

Data exports of agent chats now include the full conversation messages and tool calls.

Send an agent a link to one of your artifacts and it can pull that file straight into its sandbox to work with, as long as you have access to it.

The sidebar now shows the agent you’re viewing, with a switcher to jump to another one without going back to your agent list.

Added Qwen 3.7 Plus for agents, running under Zero Data Retention.

Organizations can now choose the models behind the Recommended, Smartest, and Fastest presets, so every agent in the org runs on the models you've picked for each tier.

Added GLM-5.2 for agents, running under Zero Data Retention.

Set evaluation criteria for an agent and have its chats automatically graded against them, so you can quickly spot which conversations went well and which need a closer look.

The agent chat composer has been rebuilt to make adding context easier. Mention your agents, skills, and apps with @, and attach files by uploading, picking an existing Gumloop file, or dragging and dropping them in.

Code sandboxes now boot-up faster and have a longer lifespan so agents can iterate on code and data within a conversation more easily.

The analytics agent can now answer questions about your triggers, skills, and the files your agents produce — like which agents have active triggers, which skills get used most, and what has been generated.

Added new open-source models to the agents.

Agents with large MCP toolsets now load tool details only when needed, keeping more context available for the task and reducing unnecessary token usage.

Enterprise teams with data export access can now choose Personal or Organization when exporting data. Personal exports include data you created or initiated, without needing to export the whole organization.

Search and navigation across Gumloop are faster and more reliable.

Skills now support Editor, Viewer, and Use Only access, so teammates can update a skill, inspect it, or use it in agents without getting broader permissions.

You can now request access to agents, skills, and Flowbooks when you hit a permission wall, and owners can approve those requests in one click.

The Gumloop MCP server now includes tools for agents, sessions, skills, files, teams, connected MCP servers, resources, prompts, and MCP tool calls.

Agents can now work with Looker dashboards, Looks, folders, scheduled plans, alerts, LookML models, and query results through MCP.

BigQuery connections now support Workload Identity Federation, so teams can connect BigQuery without long-lived service account keys.

• Google BigQuery

You can now store variables and secrets at the team level, so everyone shares the same credentials without each person entering them again.

Added Claude Opus 4.8, now available as the Smartest model for your agents.

The Apify MCP can search the Apify Store for a scraper, run it, and return the results. Use it to scrape LinkedIn profiles, pull company pages, or collect job postings for lead research and outreach.

Agents can now create HTML artifacts with separate files for scripts, styles, and assets. That makes generated dashboards, forms, and small apps easier to build without cramming everything into one file.

Agent chats now show context usage, so you can tell when a long conversation is getting close to the model limit.

Agents can now work across Freshdesk support tickets and Freshsales CRM records, from replying to customers and updating contacts to managing deals, tasks, and sales activity.

Agent reflection reports can now be sent to Slack or email, making it easier to review what an agent learned and what it wants to improve.

Deleting an agent now shows which triggers will be affected before you confirm.

Agents now support queued follow-up messages while an agent is still working, so you can add steering instructions mid-chat and have them sent in order.

Agent chats now support speech-to-text input, letting you dictate prompts directly into the composer before sending.

Subagents can now keep working for up to 1 hour, instead of the previous 15-minute execution window.

Agents with large MCP toolsets can now load tool details into context only when needed, freeing up more of the context window for the task itself and reducing unnecessary token costs.

Agents now automatically learn from past interactions. Reflections surface patterns and suggestions, helping your agents improve over time.

When someone requests a role change or team access, admins can approve or reject directly from Slack, or open the request link to see exactly what permissions would change before deciding.

Each hosted or proxied MCP tool on an agent can now be pinned to a specific personal or team credential, so you control exactly which account each tool uses.

The Home tab now shows agent activity across agents you created, agents shared with you, and agents from teams you are part of, including recent chats, usage, and triggers.

You can now favorite agents to make them easily accessible in the sidebar. If you have not picked any favorites yet, your most-used agents will appear there instead.

Agent chats now support any file type up to 200MB, with smarter routing that decides whether to send files directly to the model or make them available in the sandbox based on provider limits.

SCIM group mappings can now sync users into Gumloop teams and custom roles, including users who belong to multiple IdP groups.

Approvers now get Slack notifications when someone requests access, helping teams grant access faster.

Edit skill files directly with change previews, draft tracking, and multi-file save support.

Rename any agent conversation to keep your history organized, and share individual conversations with teammates.

Filter audit logs by user, team, IP address, or entity ID to quickly find the activity you're looking for.

Pull traffic, engagement, and conversion data directly from Google Analytics.

Manage calendar events, check availability, and schedule meetings directly from Outlook Calendar.

Enterprise organizations can now configure data drains to continuously export data to external destinations like S3, BigQuery, or your own data warehouse.

Pull search performance data, inspect URLs, and monitor indexing directly from Google Search Console.

Now available for image generation across agents & workflows.

Pro tier members can now purchase up to 1M credits via self-serve billing.

Agents can now spawn subagents and broadcast progress between siblings so multi-agent workflows stay coordinated.

A built-in analyst for your organization. Ask anything about credits, agent adoption, or MCP activity and get charts and reports back instantly.

Added GPT-5.5 support across agents and workflows.

Organization admins can now manage who has access to each team.

Agent chats can now keep running in the background so you can leave and come back while long-running work continues.

Added Claude Opus 4.7 as the default "Smartest" model.

Skills now track version history and show you exactly what changed across iterations.

Agents can now fetch and extract content from web pages directly.

Credit usage now counts up everywhere instead of counting down, with a revamped credit limit display. Configurable notification threshold controls let you set alerts for credit usage. A new Subscriptions tab in settings gives you a dedicated view of your plan.

Added Google Slides MCP integration — create, read, and modify presentations directly from your agents.

Added a native trigger that fires when a new Jira issue is created, so your agents can respond instantly.

Added a new trigger type that monitors web pages for changes in parallel.

Search on the Agents and Workflows pages now works globally across all folders — no more switching folders to find what you need.

When you paste a large block of text, it's automatically converted into a file attachment so your agent can work with it without hitting context limits.

Share agents, workflows, and workbooks with granular roles. Editors can modify, viewers can see, and use-only users can chat with agents without access to the builder.

Added Shopify MCP integration with full OAuth support.

Jira and Confluence MCP servers now support custom Atlassian domains for Government Cloud environments.

We implemented a way for agents to use tokens 2x more efficiently. This effectively makes agents half as expensive as before — same credits will go twice as far!

• Agent rate limits now align with the same mechanism as workflow rate limits.
• Free tier: 5 concurrent agent interactions.
• Pro tier: 25 concurrent agent interactions (across organization).
• Enterprise: custom limits.
• Agent interactions get queued after limit is reached.

Skills in teams are now sorted by most popular (most edited and used). You can also see the edit history and attribution of edits.

Added GPT-5.4, GPT-5.4 Mini, GPT-5.4 Nano, and GPT-5.3 Codex models for agents and workflows.

Added a max_steps setting in AI advanced settings to limit the number of steps an agent can take during execution.

Plans now default to unlimited overages. Enterprise customers can configure a custom overage cap.

Added Findymail MCP integration.

Added a toggle for compact progress mode. Agents can now be much less wordy and overwhelming while they work.

Added Nano Banana 2 support for agents and workflows.

Added a Monitoring tab in Gumstack with server health visibility.

Added a toggle to disable follow-up prompts on agents.

Added Dropbox MCP integration.

Added QuickBooks MCP integration.

Added Reducto MCP integration.

Added Hex MCP integration.

Added Outlook calendar MCP to create and update events in your Outlook calendar.

All your Agents, Workflows, and Skills now live together in the sidebar organized by Team, so it’s easy to see everything your organization is working on in one place.

You can now restrict which tools are available on custom MCP servers, the same way you can for built-in integrations.

Skills can now be downloaded directly from the UI. Save them locally or share them with your team.

Just like with Skills, Agents can now update their own instructions. They refine how they operate as they handle more tasks.

We've made adding Agents to Slack a one-click process. Just select which channel your Agent should live in and you're done.

Select either Qwen or Kimi as a model in agent chat then go to "AI Advanced Settings" on the right tab to configure OpenRouter options.

Credential management pages have been revamped to make it easier to manage your personal and workspace credentials.

Added Sonnet 4.6, Kimi K2.5, and Qwen to the model lineup. Sonnet 4.6 available everywhere while Kimi and Qwen only with Chat and Agents.

Agent tool executions now show real-time captions describing what each tool is doing—like “Searching for files” or “Sending email to team.” No more guessing what’s happening behind the scenes.

Gumloop University has a new look, new home and brand new courses covering AI fundamentals, Chat, Agents and Workflows: university.gumloop.com

Organize your agents just like you organize your workflows. Create folders to group related agents by team, project, or use case. Plus, we’ve unified the way Agents and Workflows can be navigated.

Gumloop now supports Opus 4.6 (state-of-the-art model for advanced reasoning tasks) and GPT-5.2 Codex (optimized for agents and efficient tool calling). We’ve deprecated older and less-used models for a cleaner experience.

We’re renaming Flows to Workflows as we feel that is a better name for the place where you can orchestrate agents and deterministic steps.

Manage files and folders directly in Box. Upload documents, download content, and organize your cloud storage without leaving Gumloop.

Connect to Seismic for sales enablement. Access content, manage libraries, and streamline how your team finds and shares sales materials.

Agents now automatically compact conversation history during long chats. Your agents can maintain context over extended interactions without losing track of what was discussed.

Added Sprig MCP integration for user research and feedback collection. Your agents and workflows can now interact directly with Sprig to gather insights and manage feedback workflows.

Agent chat components have been refined. Completed steps are now consolidated into a collapsible section for better visibility into what your agent is currently doing.

Watch agent conversations unfold in real-time while a workflow is running. Click into any agent node during execution to see the full thread.

Share individual agent conversations with teammates or external collaborators so others can see the full magic behind your agents.

Organization admins can now restrict which image generation models are available to their team. Set allowed and denied lists, and configure fallback models when a user's preferred model isn't permitted.

The AI model picker now includes auto-selection presets — choose Recommended, Smartest, or Fastest and let the system pick the best model for each request. Hover over any model to see speed and intelligence ratings at a glance.

Share your agents with your organization or the community by publishing them as templates. Include chat replay demos so others can see exactly how your agent handles conversations before they clone it.

New nodes for Microsoft Teams let you send messages to channels, read channel history, and trigger workflows from Teams events. Connect your workflows directly to where your team already collaborates.

Sign in with your Microsoft account. Users with Microsoft SSO can now authenticate directly without creating separate credentials.

Agents now have full feature parity with Workflows. Clone agents to create quick copies, duplicate them across workspaces, or move them to different projects — all from the same dropdown menu you already use for Workflows.

Stopping a workflow now gracefully terminates all agent interactions in progress. No more orphaned agent sessions — when you abort a pipeline, every running agent receives the stop signal immediately.

Agent creation now uses icons and has improved random name generation.

Add HTML support for Gmail send email tool in agents. Agents can now send beautiful emails.

Enable Google Tasks MCP in production.

Revamp UX for selecting API key on webhook trigger form and API keys in workspaces.

We will be adding default nodes per integration so you will not have to build basic nodes yourself.

Add timezone handling for agents to convert to UTC.