Goal for this release
The last release introduced subagents, background computer work, more Hypercards, and six new connectors.
This release brings cloud browser use, more robust Hypercards, longer-running tasks, and a more personal starting point for your work. It also expands what Hyper can do across your connected tools and improves file handling, navigation, and reliability.
Our goals were to:
- Let Hyper work on websites beyond your connected tools.
- Let you watch the cloud browser, take control, and sign in when needed.
- Make Hypercards appear more consistently and work across more connectors.
- Help you find useful tasks from your own work and connected information.
- Make it easier for new teammates to get started.
- Let longer jobs continue without a fixed limit on the number of steps.
- Improve everyday editing, scrolling, file uploads, and connector reliability.
Hyper can now use a cloud browser
Hyper can now open websites, read pages, click, type, and work across tabs in its own cloud browser.
The browser runs in the cloud, without needing your computer to execute the work. It is separate from your personal browser and does not use your existing tabs, cookies, or sign-ins.
Cloud browser use was previewed in the last release. It is now enabled for stable releases. (#1418)
Work beyond your connected tools
You can ask Hyper to work on websites that do not have a connector, or complete a task that needs a website’s interface.
For example:
Open our website and check the pricing and signup pages for broken links.
Help me update this setting in the web console.
Hyper reads the page after each step and uses the current page to decide what to do next.
Browser actions appear in the conversation with descriptions such as Open Google Cloud or Click Save. (#1241, #1299)
Watch Hyper work
A Browser chip appears above the chat bar while the conversation has an open cloud browser.
Select it to see a live preview. Open the larger view to follow the page, Hyper’s current action, and the steps it has taken.
While you watch, clicks and typing happen visibly on the page. Closing the preview does not stop the browser.
The live browser is available in the desktop app and is visible only to the conversation’s owner. (#1299, #1389, #1391, #1392)
Take control whenever you need to
Choose Take control to use the browser yourself.
You can correct a field, accept a cookie banner, navigate to the right page, or complete a step that needs your input.
Hyper pauses its browser work while you have control. Choose Release control when you are done. Hyper then reads the page again before continuing.
Taking control during a step interrupts that step rather than letting Hyper continue clicking while you work. (#1389, #1390, #1391, #1392)
Sign in when Hyper asks
When a website needs a sign-in or another step only you can complete, Hyper shows a request in the conversation.
Choose Open browser, complete the step yourself, and select Done. Hyper continues from the page as you left it.
Choose Skip if you do not want to continue.
Hyper does not type your password or verification code. On iPhone, you can see and skip the request, but completing it requires the desktop app. (#1384, #1387, #1388)
Browser sessions stay with the conversation
Each conversation has its own cloud browser. Later messages in the same conversation can use its open pages and sign-ins while that browser remains active.
The browser closes after ten minutes without a browser step. A browser waiting for your response stays open until you answer or the request expires.
Sign-ins are not saved after the browser closes and do not carry over to another conversation. Some websites may also block cloud browsers.
Browser permission remains separate from connector permissions. (#1299, #1384, #1413, #1417)
More robust Hypercards
Hypercards now appear more consistently and work more reliably across your connected tools, with expanded support for connectors such as Attio, HubSpot, and Trello.
This release improves existing cards as well as adding new ways to work with records, saved email drafts, and issue statuses.
Work with Attio and HubSpot records
Attio and HubSpot records can now appear as Hypercards.
You can view supported records and draft new ones directly in the conversation. Record drafts use the service’s field information, including supported choices and readable names.
For example:
Draft an Attio record for this company using the details from our conversation.
Show me this HubSpot record, then prepare a new record for the related opportunity.
Review and edit the available fields before creating the record.
These additions cover supported record reads and creation. They do not mean that every CRM operation has an editable card. (#1277, #1309, #1354)
Change issue statuses and move Trello cards
Supported change cards now cover more than titles and descriptions.
You can review supported status changes for GitHub, GitLab, Asana, and Linear, and move a Trello card to another list.
The available choices depend on the connected service and item. Review the proposed change before saving it. (#1276, #1306, #1310, #1315)
Start an email before every detail is ready
Gmail and Outlook compose cards can now hold incomplete emails.
You can start with the message, add the recipient later, or continue working on a draft before it is ready to send.
Saving a draft does not require it to be ready for delivery. Sending still requires the necessary fields. (#1331)
Edit and send saved drafts
Supported Gmail and Outlook draft updates now appear as compose cards.
Hyper can also show a saved Gmail draft or Outlook message for review before sending it. The card reflects the saved item rather than reconstructing an unrelated email.
Before sending a saved item, Hyper checks that it still matches what was reviewed. (#1336, #1339, #1344)
More requests stay in cards
We expanded the request shapes that existing cards can represent.
Supported email, message, calendar, booking, issue, task, and record requests are less likely to fall back to a plain-text result simply because part of the request is not editable.
This improves card coverage across existing connectors without changing the underlying service’s capabilities. (#1333, #1341, #1347, #1352, #1354)
Rich-text editing on iPhone
iPhone now uses the same rich-text editor as desktop for:
- Email bodies.
- Channel messages.
- Issue and task descriptions.
- Proposed description changes.
- Automation instructions.
You can edit formatted content instead of working with raw Markdown.
Read-only card content also preserves line breaks and displays HTML as text more consistently. (#1275)
Tasks suggested for your work
The chat bar now suggests tasks that are relevant to your work, based on your recent chats and connected tools.
Suggestions can use information such as upcoming meetings, unanswered threads, open issues, and recent changes.
For example:
Prep me for Thursday’s call with Acme.
What changed in the project since our last review?
The suggestions use information available to you in the current workspace, including searchable services shared by teammates. They do not grant access to another person’s private account. (#1404)
Start from a suggestion
Suggestions rotate in the chat bar when you open an empty new chat.
Press Tab to place the current suggestion in the composer, then edit or send it.
Hyper refreshes the suggestions in the background. If a personal list is not ready, the chat bar continues to show general examples.
Generating suggestions only reads information. It does not send messages or change connected records. (#1404, #1410, #1416)
A better start with Hyper
New users get a guided introduction and a getting-started conversation that helps them connect their tools and try useful work.
A guided introduction
Four introductory screens explain what Hyper can do before opening the app.
Your progress is saved, so leaving setup does not require starting again. You can return to the screen where you stopped, including on another device.
New users name a workspace during setup. People joining through an invitation enter the invited workspace instead.
Existing users who already completed setup do not have to repeat the introduction. (#1364, #1377, #1403, #1412)
Connect useful services sooner
The getting-started conversation suggests services that are likely to be useful to you.
Hyper can use sender information from your inbox and public information about your company to identify likely tools. Common services fill the gaps when there is not enough evidence.
The inbox scan used to identify services counts matching senders; it does not read message bodies.
Suggestions appear as connection cards that you can use when you are ready. (#1273, #1329, #1332, #1351, #1361)
Connect Gmail and Calendar during Google sign-in
Google sign-in now requests the permissions needed to connect Gmail and Google Calendar.
When granted, those connections can be attached as you join or create a workspace, instead of requiring a separate connection flow for each one.
Google Drive still connects separately.
Existing workspace memberships are not automatically backfilled with these sign-in connections. (#1312, #1321, #1401)
Start with your own work
When the available permissions and information allow it, the getting-started chat includes:
- A short introduction based on what Hyper knows about your work.
- Upcoming meetings you can ask Hyper to help prepare for.
- A work email that appears to need a reply, with a draft you can review.
The reply stays in a compose card until you choose to send it.
The welcome shows one qualifying email thread. If there is not enough information for a section, Hyper leaves it out rather than inventing content.
This is a one-time getting-started conversation for a new workspace membership, not a recurring daily briefing. (#1372, #1378, #1393, #1416)
See what your team already connected
Invited teammates see a short explanation of services their team already shares.
Their own connection suggestions remain separate, so shared company knowledge is not confused with access to a personal account.
Welcome cards stay beside the relevant part of the message. The chat opens at the beginning and keeps its position as more briefing content arrives. (#1394, #1421, #1416)
Longer-running tasks
Hyper can keep working through longer jobs without stopping after a fixed number of model or tool calls.
Previously, a run stopped after:
- 100 model calls.
- 200 tool calls.
Those fixed call limits have been removed. Longer debugging, research, and multi-step jobs can continue while Hyper manages the conversation context.
Task-duration, computer-session, and provider limits still apply. (#1289, #1256)
Better follow-through
We revised Hyper’s instructions for completing work, checking results, and responding to your feedback.
The changes focus on:
- Continuing through recoverable failures.
- Completing every item in a requested batch.
- Checking the result before reporting a task as finished.
- Following the examples and writing style you provide.
- Keeping your corrections when a long conversation is summarized.
Hyper also distinguishes more clearly between a question that needs an answer and a request that needs work completed. (#1342)
More selective use of subagents
Hyper now keeps ordinary tasks together instead of delegating small pieces unnecessarily.
Subagents remain available for substantial, independent work where parallel execution helps, or when you explicitly ask Hyper to delegate.
Completed subagents also leave the active-work area more reliably, while their activity remains available in the conversation. (#1284, #1239)
Existing connectors can do more
More Gmail actions
Gmail now supports broader email management, including:
- Archiving messages.
- Applying labels.
- Marking email as read or unread.
- Moving messages to the trash.
- Permanently deleting messages.
- Creating labels and filters.
- Setting a vacation reply.
These additions build on the existing read, draft, reply, and send capabilities.
Forwarding, send-as configuration, and mailbox delegation remain outside this access. (#1335, #1291)
More Google Calendar management
Hyper can list your calendars and work with supported calendar-management actions, including creating calendars and changing who can access them.
This expands beyond the event-management capabilities available previously.
Available operations still depend on your Google account’s permissions. (#1335)
Manage more Google Drive files
Hyper can now rename, move, copy, delete, share, and comment on files you have permission to manage, rather than limiting those actions to files Hyper created.
Editing Google Docs, Sheets, and Slides remains supported.
Existing connections retain their previous Google permissions. If a newly supported action is refused, disconnect and connect the service again to grant the expanded access. (#1335)
Broader LinkedIn and WhatsApp support
We expanded the exposed LinkedIn API and WhatsApp messaging operations.
Hyper can use more of the connected service’s supported capabilities instead of being limited by an incomplete interface.
The account’s permissions and the provider’s restrictions still determine which operations are available. (#1343, #1340)
Allow always for more actions
Connector actions that previously required approval every time can now offer Allow always.
You can choose whether to approve one action or save permission for that kind of action.
This does not automatically approve anything. Existing grants, draft requests, and your approval choices still control what runs. (#1353)
Larger PDF uploads
You can now upload PDFs up to 50 MB.
The combined upload limit has increased from 10 MiB to 50 MB per message, with the existing limit of five attachments.
This makes it easier to work with longer reports and documents without splitting them into smaller files first. (#1383)
More reliable handling of large documents
PDFs now use reusable file references with the model provider instead of being copied into every request.
Older PDFs remain available through file tools when they do not fit in the current model request.
Images that exceed the inline image budget also remain available through file tools.
Connector downloads, extracted Office content, and generated computer outputs retain their existing limits. (#1383, #1294)
A new privacy control
Under Settings → Privacy, you can choose whether diagnostic session recordings include conversation text and images for the Hyper team.
The setting is named Securely share my conversations with the Hyper team and is available on desktop and iPhone.
New installations default to content sharing being off. Existing installations keep saved preferences or the earlier content-sharing default.
The choice is saved on the device and applies to future recording content without restarting the app. (#1338)
Usage information and task processing are separate
This setting does not turn off basic usage information or the processing needed to complete a task you request.
On desktop, turning it off masks conversation text and images in diagnostic recordings.
The setting is separate from microphone transcription and from the information sent to an AI provider to answer your request. (#1338)
Dictation opts out of model improvement
Desktop dictation now explicitly opts out of Deepgram’s model-improvement program when it starts transcription.
This does not change how you record, review, or send dictated text. (#1360)
Ask Hyper how to use Hyper
Hyper now has a searchable product manual that it can consult when you ask about features, settings, connectors, or troubleshooting.
For example:
How do I change an automation’s instructions?
What can Hyper access through this connector?
Why is this task asking me to choose a folder?
The manual covers desktop and iPhone behavior, account setup, workspaces, files, approvals, automations, and current limits.
Connector guidance uses the same public capability descriptions shown in Settings. (#1337)
A smoother conversation view
Scroll continuously between replies
Conversations now scroll more naturally between earlier and later turns.
Short replies no longer reserve a full screen of empty space. The navigation rail follows the turn at the top of the view, making movement between short exchanges more predictable.
User messages stay in the conversation rather than appearing in floating overlays. (#1348, #1407, #1408)
Smoother mouse-wheel movement
Mouse-wheel steps now ease into position instead of moving abruptly.
Precise trackpad scrolling remains native, and reduced-motion preferences are respected.
We also improved handling of accelerated mouse-wheel input on macOS. (#1409, #1422)
Cleaner message controls
Controls for your messages now sit beside the message bubble.
We also fixed welcome conversations opening at the bottom before their introductory content had finished loading. Ordinary conversations still open at the latest message. (#1406, #1416)
Performance and reliability fixes
Granola preserves stored transcripts
Granola synchronization now handles Basic and free-plan limitations more reliably.
Hyper reads the available recent meetings, uses the meeting’s actual start time, and preserves transcripts already stored when a later response contains less information.
A free-plan sync no longer removes a previously stored transcript simply because Granola does not return it again.
This preserves available content; it does not bypass Granola’s plan restrictions. (#1323, #1328, #1350)
Cleaner HubSpot knowledge
HubSpot synchronization now focuses on useful work records and recent messages.
We reduced unnecessary indexed content, limited message indexing to the last month, removed internal fields, and skipped empty messages and notes.
These changes make the indexed information more useful and reduce unnecessary synchronization work. (#1280, #1298, #1346, #1362)
More reliable LinkedIn updates
A refused later page of LinkedIn posts no longer fails the entire synchronization.
New connection events also refresh the relevant relationship rather than rereading the full relationship list.
Other available information can continue to update when one part of LinkedIn’s API is unavailable. (#1279, #1368)
Better Google Drive and Vapi synchronization
- Google Drive: Files too large for Google to export are skipped instead of stopping the synchronization.
- Vapi: Empty strings returned in place of optional objects no longer cause the response to fail parsing.
An oversized Drive file that was skipped is not included in Hyper’s synced copy. (#1325, #1318)
More reliable Webflow connections
Hyper can now be installed through Webflow’s marketplace-listing flow.
The installation is tied to the browser session that started it, and revoking a Webflow connection stops further calls and synchronization through that connection. (#1313, #1327, #1334)
Clearer errors and delivery status
Error messages now explain more precisely what went wrong and what you can do next.
Attachment errors distinguish file size, file count, image dimensions, unsupported content, and processing limits.
Interrupted or timed-out sends no longer claim that nothing was sent when the outcome is unknown. Hyper instead tells you to check before repeating the action. (#1375)
Better search and conversation context
Knowledge search now uses a workspace-specific index to reduce unnecessary scanning.
When Hyper searches past conversations, results identify who wrote each message more reliably. This helps distinguish your instructions and corrections from an earlier generated answer.
Each run’s account and workspace context is also preserved in history rather than being replaced with the current context when old messages are loaded. (#1285, #1314, #1395)
More consistent email and message content
Gmail text parts now reach Hyper as readable text.
We also improved Markdown conversion for card content, including Slack and Discord messages, so formatting and line breaks survive more consistently. (#1376, #1272, #1322)
Several improvements in these notes already shipped in 1.10.1–1.10.9, including longer-running tasks, larger PDF uploads, and connector fixes. They are included for anyone upgrading from 1.10.0.
Other additions and improvements
- More reliable chat previews: Older tool results are counted as part of the turn they belong to when previews are prepared. (#1311)
- Connection file counts: Hyper now keeps a file count for each connection without repeatedly counting its indexed files. (#1308)
- Existing private workspaces remain: New users create or join shared workspaces. This does not delete existing private workspaces or their contents. (#1403)
- More reliable suggestions: Fixed a background suggestion turn ending with an empty response after moving to Luna. The generated suggestions remain separate from the visible chat. (#1410, #1416)
- Updated policies: Rewrote the published Privacy Policy and Terms of Service to describe the current product. (#1359, #1365)
Engineering and release improvements
Less overhead when using connectors
Connected services now supply their request shapes directly to Hyper, and the first execution response includes the relevant connector guide.
Later searches avoid repeating guides that are already present.
We also added checks that connector guides agree with their implemented bindings.
These changes reduce repeated discovery work and give Hyper clearer instructions for using the connected service. (#1363, #1367, #1371, #1399, #1402, #1405)
Separate resources for synchronization
Connector synchronization now uses a separate database pool.
This reduces competition between background synchronization and the work needed to serve conversations.
Indexed path-prefix queries also make it cheaper to identify files that should remain in a synced collection. (#1303, #1366)
Better records for background commands
The server now saves how a background desktop command ended, including its exit code and the end of its log.
Background commands use the same underlying task records as other background work.
This is groundwork for clearer background-task controls. It does not yet add visible command chips or a unified stop control. (#1245)
Smaller completed-task storage
Completed execution records can now be removed after their results have been collected.
Conversation history, cards, submitted inputs, and automation results remain available. Subagents and their parents are retained while their relationship is still needed.
This reduces retained execution state without deleting the completed work from the conversation. (#1374)
A simpler cloud-browser implementation
We removed the unused desktop Chrome execution path now that the browser runs in the cloud.
Each cloud browser owns one window, while continuing to support multiple tabs. Browser startup, steps, control handoff, and shutdown now have clearer internal boundaries.
We also improved the expanded browser view’s opening animation and input handling, and added clearer records of why a browser closed. (#1411, #1413, #1414, #1415)
Consistent models for background work
Chat titles, the sign-in briefing, and task suggestions now use GPT-6 Luna, alongside the existing Luna-powered chat previews.
Subagent model selection is unchanged: OpenAI tasks use Luna, and Anthropic tasks use Haiku.
This changes background model routing, not the model selected for your main conversation. (#1410)
Better visibility into performance and cost
We added more detailed measurements for:
- Time spent opening a conversation.
- Fetching, applying, and rendering its transcript.
- Cloud browser and sandbox runtime.
- Compaction calls whose summaries were not applied.
- Server starts, crashes, memory, and disk usage.
Monitoring boards now make these measurements easier to compare across releases, with clearer alerts about problems users experience. (#1282, #1286, #1292, #1300, #1302, #1304, #1380, #1385, #1397)
Security and development improvements
- Database connections: Deployed environments now require TLS for PostgreSQL connections. (#1357)
- Google Calendar development access: Added the settings scope for local and nightly use; this does not add that scope to stable consent. (#1355)
- Review recordings: Improved connector demonstration recordings, including separate clips and clearer captions. (#1297)
- Website discovery: Added a sitemap and
robots.txt. (#1293)
Plans stay separate from released features
We updated the Hypercards and onboarding plans alongside their implementation.
Universal cards for every connector operation, additional discovery moments, and other planned follow-ups are not included simply because they appear in those documents.
This release includes the specific card, suggestion, and getting-started capabilities described above. (#1319, #1330, #1307, #1358, #1369, #1400)
Coming next
- A massive product revamp: We’re revamping the entire product to reflect our new branding, including the interface and Hypercards.
Have a feature your team needs? Tell us at founders@heyhyper.ai.