Every Hyper release on both channels. Stable releases are cut deliberately and carry curated notes; nightly builds ship daily with notes generated from the merged pull requests.
Full changelog: v1.8.2...v1.8.3
Full changelog: v1.8.1...v1.8.2
Full changelog: v1.8.0...v1.8.1
v1.8.0
The last release made Hyper’s agent substantially more capable: you could redirect it while it worked, queue follow-up requests, and continue the same work across devices.
This release expands the work Hyper can complete beyond connected APIs. It brings desktop and cloud computer use, Hyper for Windows, eleven new connectors, stronger file handling, and a much faster and more responsive experience.
Our goals were to:
Hyper can now run scripts to complete work that connected APIs cannot do on their own.
You can ask it to:
Computer execution works alongside Hyper’s existing search, connectors, and conversation history. A task can retrieve information from a connected service, process it with a script, and use the result in a later connector action. (#823, #824, #871)
When a desktop is connected, commands run on that computer and can use its installed tools and environment.
Without a connected desktop, Hyper uses a temporary cloud sandbox. This also lets you request computer work from your phone without making the phone itself run the commands.
Desktop files persist in the work folder. Cloud workspace files are temporary; results returned to the conversation can be previewed, downloaded, or used in another supported action. (#816, #824, #871)
Under Settings → Account → Computer, choose the folder where desktop commands should start.
You can select a repository or another existing folder, enter its path directly, or reset to Hyper’s default workspace folder. The choice stays on that computer.
The work folder is a starting directory, not a security sandbox. Desktop commands run with your local account’s permissions. (#836)
Scripts that change or delete existing files, install software, or run third-party command-line tools ask for approval before execution.
Read-only inspection, calculations, and creating new files to return can run without an extra approval prompt.
Approving a script does not approve a later action in Gmail, Slack, or another connected product. Connector actions keep their own permission checks. (#835)
Script activity now appears as terminal-style rows with live output instead of remaining silent until a command finishes.
Results identify whether the command ran on a desktop or in the cloud. Progress lines update in place, and output is bounded so a noisy command does not overwhelm the conversation.
Independent desktop commands can run together. Desktop commands also support longer execution times, up to ten minutes, and use the login shell’s tool paths. (#838, #844, #845, #846, #849, #874)
Hyper can now work with files returned by connected services and scripts, not only files you upload yourself.
For example:
Export the data, analyze it, and send me a spreadsheet with the results.
Read this document, create a revised version, and upload it to the project.
The same file can move from a connector into a script, back into the conversation, and onward to another supported connector without the model reconstructing its contents.
Hyper can inspect returned images, PDFs, text, and supported Office documents. Other formats can be processed with computer execution. (#789, #790, #791, #793)
File handling now covers more supported operations across email, storage, messaging, project-management, and CRM connectors.
This includes provider-specific handling for email attachments, multipart uploads, raw file transfers, and exported content. Hyper supplies the original file bytes in the format the destination requires.
Support remains specific to each provider and operation. Arbitrary download hosts, large resumable uploads, and some separate signed-upload procedures are not included. (#794, #795)
Uploaded and generated files now use the same authenticated preview and download path.
Images open in a lightbox, PDFs open in the document viewer, and other files retain download controls. Desktop and iPhone can save returned files.
We also fixed previews reloading during streaming, empty-file handling, and exported files being identified with the wrong type. (#702, #750, #809, #812)
You can now use Hyper on your Windows PC!
The desktop app now has a Windows installer, platform-specific release artifacts, and an update path. The download page offers the Windows build with Windows installation instructions.
We also fixed Windows session persistence and stopped the desktop app from opening an unwanted console window.
Download Hyper at heyhyper.ai/download.
The initial installer is unsigned, so Windows may show a SmartScreen warning. (#740, #747, #859, #868)
Hyper now connects to Monaco, Microsoft Planner, Sunsama, Outlook Contacts, GitBook, Gusto, Canvas, Webflow, Persona, Vapi, and Windsor.ai.
Supported operations remain limited by the connected account’s permissions, provider plans, and Hyper’s action controls.
Connect Monaco to work with accounts, contacts, opportunities, tasks, meetings, and sequence templates.
Hyper indexes CRM records using your organization’s field labels and pipeline stages, alongside available meeting notes, summaries, and transcripts.
Email and LinkedIn conversations are read on demand through Monaco rather than copied into the background index. (#741)
One connector brings together Microsoft Planner and Microsoft To Do.
Hyper indexes tasks with their plan or list, descriptions, checklists, links, and available dates. To Do changes use Microsoft’s change feed and notifications; Planner uses periodic checks.
The initial interface excludes Planner write operations that require unsupported version headers. Other supported task operations remain available through the connector. (#729)
Connect Sunsama to work with tasks, daily plans, weekly objectives, and backlog items.
Indexed tasks include notes, subtasks, priorities, estimates, time spent, due dates, and linked work. Private tasks and tasks in personal channels remain private in Hyper.
Sunsama’s supported planning and task actions are also available through its connected interface. (#760)
Outlook Contacts is a separate connection for your Microsoft address book.
Hyper indexes names, companies, job titles, email addresses, phone numbers, postal addresses, categories, and notes across the default Contacts folder and supported contact folders.
Contacts remain private. Incremental updates and notifications keep changed records current. (#761)
Connect GitBook to search collections, spaces, and page contents alongside your other company knowledge.
Hyper reads the Markdown of pages in changed spaces and preserves the distinction between deleted content and content the account can no longer access.
The connector also exposes supported GitBook API operations available to the authorized account. (#781)
Connect Gusto to work with the payroll and company information available to your account through its supported API.
Gusto is an on-demand connector: Hyper reads current records when a task needs them rather than copying payroll data into a background index.
Connections start private because they can contain salary and personal information. (#797)
Connect your school’s Canvas account using its site address and a personal access token.
Hyper indexes current coursework, including:
The connector also supports Canvas REST and GraphQL operations available to your account. (#798)
Connect Webflow to work with sites, page text, CMS collections and items, form submissions, comments, products, and orders.
Incremental synchronization and webhooks keep supported content current. Hyper can also use Webflow’s supported site and content-management actions through its Data API. (#796)
Connect Persona with an API key to work with identity-verification inquiries, accounts, cases, reports, transactions, and related resources.
Persona is read on demand rather than indexed in the background. The key’s permissions determine which operations are available.
Production and sandbox environments can be connected separately. (#818)
Connect Vapi with a private API key to work with voice assistants, calls, chats, phone numbers, and related resources.
Hyper indexes assistants, squads, phone numbers, call summaries and transcripts, and chat conversations. Supported live operations cover the broader voice-agent interface, including available recordings and analytics. (#819)
Connect Windsor.ai to query the marketing, advertising, analytics, commerce, CRM, and warehouse sources already connected to your Windsor account.
Hyper can work with metrics across channels and use supported actions such as campaign changes and scheduled exports.
Windsor is an on-demand connector. Results are queried for the dates, accounts, and fields a task needs rather than copied into a document index. (#827)
You no longer have to leave a conversation to start connecting a service.
Ask something like:
Connect my GitBook account.
Hyper can show a Connect card in the conversation. Complete the provider’s sign-in flow, then continue the task.
Existing connections use the corresponding reconnect flow. Skip remains available while sign-in is waiting in the browser, so abandoning a connection attempt does not leave the conversation stuck. (#738, #782)
We upgraded:
Existing saved selections move to their replacement models without requiring you to choose again. GPT-6 Astra remains the default. (#774, #775)
Hyper now requests fast-mode processing for supported models automatically.
Catalog GPT models use OpenAI’s priority processing, and Claude Opus 5.5 uses Anthropic’s fast mode. Other Claude models continue using standard processing where fast mode is not supported.
Actual speed still depends on the provider and task. (#850)
Hyper is also much faster and feels more responsive across the board.
Tool names and arguments appear while the model prepares them. Available reasoning summaries feed the waiting indicator, and longer waits show updated wording and elapsed time instead of leaving the same initial message on screen.
Together with live script output, these changes make it much easier to tell what Hyper is doing before the final answer arrives. (#814, #825, #847, #864, #869)
Hyper now summarizes earlier messages as a conversation fills up. This is called compaction.
After a completed reply reaches the model’s compaction threshold, Hyper saves a summary that later turns use in place of the earlier message history.
You can also request it explicitly:
Compact this conversation.
Desktop provides a Summarize now action as well. Hyper shows when summarization is running.
If compaction fails or is stopped after a completed reply, that reply is preserved. The summary does not have to succeed for you to keep the answer already produced. (#865, #866, #872, #873)
Automation setup now uses an editable card for the name, schedule, and instructions.
After editing a proposal, choose Review changes. Hyper reviews the changes and replaces the proposal before you confirm it. Editing the card does not activate an automation or bypass confirmation.
The iPhone prompt editor uses the same review path. (#828)
Owners can edit an existing automation’s name, schedule, time zone, and instructions from its settings.
Saving instructions creates a new version for future runs. Previous runs retain the instructions they used. Schedule changes recalculate the next run without resuming a paused automation.
Event triggers remain read-only in this editor. Time fields and searchable time zones make scheduled automations easier to adjust. (#828, #856, #858)
The sidebar now separates:
Selecting a Scheduled entry opens its settings, next-run information, and Run now control. Previous results remain available in the run history, with clearer day-and-time labels.
Opening settings does not mark unread results as read. (#802, #821, #826, #831)
Feedback on an automation’s output can now prompt an instruction-change proposal without requiring you to use a specific phrase such as “from now on.”
The proposed change still needs review and approval. Commenting on a result does not silently rewrite the automation’s instructions. (#800)
Under Settings → Responses → Danger, you can enable YOLO mode to let Hyper perform connector actions without approval prompts during your interactive conversations.
YOLO mode is off by default.
While enabled, it can allow sends, updates, and deletions without another confirmation. Use it only when you are comfortable letting Hyper take those actions directly.
It does not change:
Turning it off restores normal connector approval rules without leaving behind permanent approval grants. (#716)
Hyper now sends desktop notifications when a live response finishes or a conversation needs your input.
Notifications identify the conversation and show whether an answer, question, or approval is waiting. Clicking one opens the relevant chat.
Closing the main window now keeps Hyper running. Reopen it from the macOS Dock or the Windows tray. Explicitly quitting stops local notification delivery.
Alerts are suppressed while Hyper is focused, and loading old conversations does not generate new completion notifications. (#772)
Concurrent knowledge searches could wait for extra database connections while already holding connections themselves, eventually timing out.
Each file and chat query now acquires and releases its own connection. A failure in one source also no longer discards successful results from the other. (#833)
Connectors now respect provider retry times expressed as dates, account for differences between clocks, and avoid retrying too aggressively when a provider keeps refusing requests.
Synchronization also holds database locks for shorter periods and coordinates account-level pauses more consistently. (#783, #785, #799, #801, #804)
(#765, #776, #778, #779, #786, #815)
The Anthropic strict-tool grammar fix also shipped in 1.7.1 and is included for anyone upgrading from 1.7.0. (#822)
We removed the old agent executor, separated release tooling into its own library, and expanded platform-specific build and publishing support.
Connector changes now trigger broader interface validation. Additional timing measurements help distinguish model latency, tool execution, and the wait before visible output. Model usage is also recorded per agent call for more accurate cost analysis.
The stable release-tool correction already shipped in 1.7.1. (#726, #744, #759, #834, #840, #852, #860)
Firebase, BigQuery, Google Analytics, Google Ads, and Google Search Console are implemented and awaiting approval.
These are separate from the eleven new connectors available in this release.
Have a feature your team needs? Tell us at founders@heyhyper.ai.
Full changelog: v1.7.0...v1.7.1
v1.7.0
This release brings a far more capable agent, eleven new connectors, conversation sharing, voice dictation, and Google Calendar, Sheets, and Slides editing.
Our goals were to:
Every Hyper app now uses the new agent runtime.
The agent runs on the server and saves progress between operations. Closing the app or losing the connection does not cancel accepted work. You can return to the conversation to review its progress and result. (#597, #613, #679)
You no longer have to wait for a reply to finish before adding something important.
Send another message during a task to clarify the request, add context, or change direction. The agent incorporates it between operations rather than restarting the entire task. Calls already in progress can finish before the new instruction takes effect.
Your message can include text, pasted content, attachments, and comments on earlier answers. (#617, #618)
You can also queue a follow-up to run after the current reply.
On desktop:
Queued messages can be edited, removed, or sent immediately while they are still waiting. Once accepted, they are stored on the server rather than depending on a local queue in the app. Editing or removing a waiting message cancels its saved entry before replacing it.
Delivery tracking also keeps an uncertain send visible instead of silently losing it or submitting another copy. (#622, #669)
When a task needs several independent reads or actions, the agent can run up to four tool calls together instead of waiting for each one in sequence.
Completed results are saved separately. If you redirect the task, the agent stops starting additional calls from the old plan, lets active calls finish, and then applies your instruction. (#648)
Questions, connector approvals, confirmations, and reconnect prompts now save the pending work instead of holding an open execution session.
You can return later, answer the card, and continue from the saved step. These cards no longer expire simply because time has passed, though stopping the task, changing the request, or changing the relevant authorization can still invalidate them.
Existing approval requirements remain in place. Saving a task does not give it additional permission to act. (#671)
Stopping a task now cancels server-side execution while keeping the partial reply and tool results already received.
You can review that work and follow up without starting from an empty conversation. An action already sent to another service may still complete; stopping the agent does not undo external changes. (#633, #662)
Chat tasks and automation runs now use durable jobs with saved progress and explicit ownership of each running job.
After an interruption, the server can recover a saved model step or pending interaction. Tool calls whose outcomes are uncertain are not blindly repeated. This avoids treating a lost response as permission to send the same external action again. (#608, #686)
Open a conversation on another device to see its active work and pending questions or approvals.
The chat list reflects work started elsewhere, including new conversations whose first response is still running. Saved messages refresh during an active task instead of appearing only after it finishes. (#630, #631, #632)
Scheduled and event-triggered automations use the same new agent runtime. Existing automation definitions do not need to be recreated, and their confirmed instructions and action permissions remain in effect. (#668, #679)
Hyper now connects to Outlook, Outlook Calendar, OneDrive, SharePoint, Microsoft Teams, OneNote, Asana, GitLab, Zoom, Loom, and Discord.
Supported actions follow each connector’s permissions and approval requirements. Provider plans, account roles, and organization settings can limit which features are available.
Connect a personal, work, or school Microsoft account to:
Mail is indexed with incremental updates and change notifications. Junk and deleted mail are excluded from indexing, and Outlook connections start private.
The initial attachment path supports files within Microsoft Graph’s inline attachment limit; larger uploads require a separate upload-session path. (#625)
Outlook Calendar is a separate connection for:
Calendar synchronization follows Microsoft’s change feed. Events marked private or otherwise sensitive remain private in Hyper. (#657)
Connect a personal, work, or school OneDrive account to find files and use supported file actions.
Hyper indexes the contents of supported Word documents, Excel workbooks, PowerPoint presentations, HTML, and plain-text files. Other supported file records remain searchable by name and folder.
You can also upload a chat attachment or replace an existing file’s contents through supported actions.
Files available to anyone with the link or to the owner’s organization can be shared with the Hyper workspace, subject to the connection’s sharing setting. Other files remain private.
This initial version does not index PDF contents or files available only through “Shared with me.” (#695)
Connect a work or school Microsoft account to search:
File and list updates use change feeds, while periodic reconciliation catches changes those feeds do not cover. Unpublished page drafts are not copied into the shared index.
The connector also exposes supported actions across SharePoint sites, lists, pages, and files. As with OneDrive, PDF contents are not indexed yet. (#674)
Connect Microsoft Teams to search channel threads, chats, and available meeting transcripts.
Standard channel threads can be shared with the workspace. Private and shared channels, chats, and meeting transcripts remain private in Hyper.
The connector also supports Microsoft Graph operations for teams, chats, meetings, and related resources.
Teams requires a work or school account and organization administrator consent for the requested channel-message access. Transcript availability also depends on the organization’s settings. (#696)
Connect a personal, work, or school Microsoft account to search the text of your OneNote pages, with their notebook, section group, and section.
Pages in a notebook you have shared can be shared with the workspace. Pages in other notebooks remain private, and OneNote connections start private.
Synchronization checks for changed sections and reads only the pages that changed. Page deletions are picked up during full reconciliation.
The agent can read notebooks, sections, pages, and page content. Creating new pages is not included in this release. (#715)
Connect Asana to search:
Incremental synchronization and webhooks keep changed records current. Where an Asana plan does not include task search, synchronization falls back to project task listings.
Supported API actions let the agent update work in Asana, subject to your permissions and approval. (#661)
Connect GitLab.com to search issues, merge requests, comments, and wiki pages from projects you belong to.
Pasted GitLab links can resolve directly to indexed issues, merge requests, and wiki pages. Confidential issues remain private.
The connector exposes supported GitLab API operations and uses project webhooks where the connected account has sufficient access.
Self-hosted GitLab is not included in this release. (#655)
Connect Zoom to search hosted meetings with available summaries, cloud recordings, and transcripts.
The connector brings together the information available for each meeting, rather than treating its summary and recording as unrelated records. Meetings without a summary or recording are not included in this initial index.
The agent can also use supported Zoom Workplace operations across meetings, chat, calendar, and other services available to the connected account. Account-admin-only operations are not included. (#626)
Connect Loom to search owned and shared videos, including available:
The connector uses Loom’s tools through Atlassian’s interface. Supported writes require approval for each call. Transcript and download availability depend on the Loom plan and the provider’s permissions. (#627)
Install Hyper’s bot in a Discord server to search accessible channels and threads and use supported messaging actions.
Each connection is bound to one server. Requests for channels in another server are refused, even if the bot is installed there too.
Discord writes require approval for each request and remain limited by the bot’s role permissions. Synchronization uses polling; edits and deletions are picked up during full reconciliation rather than through live message events. (#628)
Use Share → Create link in a conversation to create a read-only copy that someone else can open in their browser.
The shared copy includes the conversation’s messages, tool activity, and comments. Questions and approval cards show their saved state without controls for the reader to act on them.
Each link captures the conversation when you share it. Later messages do not update that copy; create another link to share a newer version.
Anyone with the link can read the shared conversation without signing in. Review the contents before sharing sensitive information. (#675, #693, #699)
Shared-chat pages follow the reader’s light or dark appearance setting. Share tokens are removed from analytics URLs, and session replay is disabled on shared conversations. (#713)
You can now dictate requests in the desktop composer.
Click the microphone and speak. A live transcription and input-level indicator show what is being captured. Choose Done to insert the transcript into your draft, then review or edit it before sending. Cancel discards the dictation.
Audio is transcribed through Deepgram. (#638)
Under Settings → Account → Dictation, choose the microphone the app should use.
The preference stays on that computer. If the selected microphone is disconnected, dictation falls back to the system default rather than failing. (#666)
The Google editing capabilities previewed in the previous release are now enabled for stable builds.
You can ask the agent to:
Reconnect Google Calendar and Google Drive to grant the new permissions. Existing connections do not receive the additional access automatically.
Editing remains subject to the applicable approval requirements. (#619)
We expanded the permissions requested by Slack, Confluence, Attio, Supabase, Airtable, Krisp, Logfire, and PostHog.
This exposes more of the operations supported by each service. For example, Slack’s expanded interface includes additional operations for files, pins, reactions, reminders, conversations, and profiles.
The expanded permissions do not override provider plan restrictions or administrator-only access. Actions remain subject to their approval rules.
Reconnect existing accounts to grant the additional permissions. (#629, #644)
When a provider refuses a connection during a task, the agent can now show a Reconnect card at the failed request.
The card identifies the affected account and offers:
The retry uses the renewed authorization for the same account. This recovery card is for interactive conversations; unattended automations do not assume someone is present to sign in. (#659)
Conversations with at least two sent user messages now show a navigation rail beside the transcript.
Hover over a marker to see its message label, then select it to jump directly to that turn. The rail supports keyboard navigation and scrolling with a mouse wheel or trackpad.
After a jump, the conversation holds your selected reading position while new output arrives. Returning to the newest message resumes following the live reply. (#652, #681)
We fixed a desktop issue that caused memory use to grow rapidly while replies streamed and remain high afterward.
The app now sends small state-change notices and retrieves the current state when needed, rather than embedding the entire conversation in every update event. This removes the repeated retention of large chat payloads without changing how replies appear. (#724)
Loading connection status no longer joins every indexed file into a large intermediate result just to count files per connection.
In the production query-plan check recorded with the change, a workspace with 78,535 files went from roughly four seconds to 42 milliseconds. Exact file counts and other status information are preserved. (#720)
A Linear issue with more than 100 comments, or a GitHub record with additional pages of nested results, could repeatedly stop synchronization.
Hyper now follows the remaining pages before publishing the record. This covers:
Records are stored only after the required pages are complete, rather than silently publishing partial content. (#723)
, or from the Hyper menu on macOS. (#665)The iOS composer-state fix also shipped in 1.6.1 and is included for anyone upgrading from 1.6.0. (#603)
We separated agent execution, connector services, and shared interaction cards into dedicated libraries. The server still owns application-specific APIs and conversation storage, while the extracted components keep their existing authorization, saved-state, and transaction contracts. (#717)
We also reduced the command-line tool’s cold build time. In the recorded M5 measurements, a fresh build went from 3 minutes 52 seconds to 58 seconds by removing unnecessary dependencies from the default build and compiling specialized commands only when needed. (#719)
These changes support faster development and releases; they do not change the permissions granted to the agent.
Have a feature your team needs? Tell us at founders@heyhyper.ai.
Full changelog: v1.6.0...v1.6.1
v1.6.0
This is a lighter release focused on stabilization and streamlining after last week’s automations launch. We’ve made automations easier to navigate, expanded what Hyper can do with attached files, and fixed issues affecting Gmail and Anthropic models.
Meanwhile, we’re preparing bigger updates for the next few weeks: cloud computer use, substantially expanded agent capabilities, and interfaces generated inline for the task at hand.
Each automation now has one row in the sidebar, keeping its results together instead of mixing individual runs into your normal chat list.
Open an automation to see its latest run. A run list alongside the conversation lets you browse earlier results, newest first, without losing your place. An unread indicator on the automation’s sidebar row helps you spot results you haven’t reviewed yet.
If an automation hasn’t run, you’ll see an overview with its trigger, next run or current status, and a Run now button. Starting a run opens its conversation as soon as it’s created.
Owners can also run, pause, resume, or delete an automation from its sidebar menu. Back and forward navigation preserves the automation view, while tab-switching shortcuts move between its runs. (#586)
Hyper could already read files attached to a conversation. Now, supported connector actions can use those original files too.
You can ask Hyper to:
Email this PDF to the customer.
Upload this presentation to Google Drive.
Hyper passes the original attachment to the connector rather than trying to reconstruct it from extracted text. This preserves the file itself, not just the information Hyper read from it.
Initial support covers:
File actions follow the applicable connector permissions and approval requirements. Hyper checks access to the conversation and validates file counts and sizes before sending.
This is the first stage of broader file support; more connector upload paths are coming. Previously attached Office files need to be reattached when an action requires their original bytes. Existing Confluence connections may also need to reconnect to grant attachment-write permission. (#598)
= to zoom in, - to zoom out, and 0 to return to 100%. Zoom resets when the app restarts. (#583)The search, Gmail, and Anthropic improvements also shipped in 1.5.1 and are included for anyone upgrading from 1.5.0.
Over the next few weeks, we’re preparing larger updates across three areas:
We’re also working on broader Google Calendar, Sheets, and Slides editing, including managing calendar events and editing existing spreadsheets and presentations. (#584)
This week’s release makes the current experience smoother while we build toward those larger capabilities.
Have a feature your team needs? Tell us at founders@heyhyper.ai.
Full changelog: v1.5.0...v1.5.1
v1.5.0
The last release rebuilt Hyper’s connector system, expanded what the agent could do across connected products, and brought Hyper to iPhone.
It also made the next limitation clear: Hyper could do substantial work when someone was actively chatting with it, but it could not continue doing recurring work after they left. Connector synchronization also needed to become faster and more reliable as the number of supported products and connected accounts grew.
Our goal for this release was to address those problems directly:
We shipped 130 pull requests across these areas.
Hyper can now continue doing work after you close the app.
Start a new chat and describe something you want Hyper to do repeatedly. For example:
Every weekday morning, summarize the new emails that need a reply.
Every Friday, review our open Linear issues and write a project update.
When a new meeting appears on my calendar, prepare a briefing using the attendee, company, and previous conversations.
Hyper helps turn the request into an automation, asks a blocking question only when something important is missing, and shows a confirmation card before anything starts. The card states the instructions, trigger, connected accounts, sharing, and any external actions the automation may take. Nothing runs until the owner confirms it. (#515, #525, #552)
An automation can run:
Schedules use the person’s actual IANA timezone rather than a fixed UTC offset. This keeps the local time correct through daylight-saving changes, travel, short months, and other calendar boundaries. A time skipped by a spring-forward transition moves to the next valid minute, while a repeated fall-back time runs only once. (#481, #522, #549)
An automation can also run when Hyper receives:
For example, an automation can watch for a new finance email, file the invoice, and prepare a reply.
Event automations begin watching from the time they are created or resumed. They do not run against an old mailbox, calendar, or issue backlog. Each connected item can create at most one run, even if the provider delivers the same event more than once.
Burst controls prevent a busy source from flooding the chat list. An automation can run up to 12 times per hour from new items; additional occurrences are recorded as missed rather than queued without limit. (#532)
Each run appears as a normal conversation in Hyper.
The conversation shows:
New runs appear with an unread indicator and synchronize across devices. Opening the conversation marks it as read for that person without changing its state for other workspace members. You can continue the conversation normally to ask a follow-up or refine the result. (#509, #550, #565)
Scheduled work is claimed transactionally, so one occurrence creates at most one run conversation. If a worker stops unexpectedly, Hyper can reclaim the existing run instead of creating another copy. Runs use durable leases, bounded retries, and explicit delayed and missed states. (#508, #522, #551)
An unattended automation should not inherit every permission its owner has granted during unrelated conversations.
The confirmation card therefore lists every external action the automation may take, including the product, connected account, effect, and destination. The owner chooses how each action behaves:
An automation receives authority only for the actions confirmed on its own card. An earlier “always allow” answer from a normal conversation cannot silently expand what the automation is allowed to do. (#531)
When a run encounters an action that still needs approval, it parks instead of holding an open process. The approval remains in the run conversation. After the owner decides, Hyper resumes from the stored tool call and avoids executing an already completed action twice.
If an external action was sent but its result could not be recorded, Hyper reports that the outcome is unknown instead of retrying blindly. (#528)
You can ask Hyper to:
Hyper lists the current automations before acting and uses their exact IDs internally. If several automations could match a request, it asks which one you mean instead of guessing.
The automation list shows its trigger, next run, current state, owner, sharing, and whether it is running now. The same lifecycle controls are available from the list. (#490, #526)
Each automation also keeps its selected model. Scheduled, event-triggered, resumed, and manually started runs all use that model rather than silently falling back to the current product default. (#527)
Feedback in a run conversation can apply only to the current result or become a proposed change to future instructions.
Hyper distinguishes between requests such as:
Rewrite this result with a shorter introduction.
and:
From now on, leave the introduction out.
When the intended scope is unclear, Hyper asks whether the feedback applies only to this run or to future runs too.
A future-facing change appears as a line-by-line instruction diff. The owner can apply or reject it. Applied changes create a new immutable instruction version, while previous runs remain attached to the instructions they actually used. Changes can also be undone without rewriting that history. (#529)
An automation can remain private or be shared with the workspace.
Shared automation runs are readable by workspace members, but only the owner can:
Runs continue using the owner’s connected accounts and permissions. If sharing could expose information from one of the owner’s private connectors, Hyper warns the owner and requires an explicit acknowledgement before enabling workspace sharing.
If the owner leaves the workspace, their automations pause automatically rather than continuing under an identity that is no longer a member. (#530)
A failed or limited run now leaves a note in its conversation explaining:
A reply produced without access to a required source is marked as completed with limitations rather than presented as fully complete.
After three consecutive failed runs, Hyper moves the automation into Needs attention and stops scheduling more occurrences. The owner can repair the underlying problem and resume it. Missed occurrences do not count as failures because no work ran. (#524)
The iPhone app can now show automation proposals, lists, runs, instruction changes, approvals, and sharing choices.
Owners can choose Draft or Automatic for proposed actions, make an automation private or shared, run or pause it, repair failures, delete it with confirmation, and apply or reject instruction changes. Shared automations remain read-only for members who do not own them. (#554, #558)
GPT-6 Astra is now Hyper’s default model.
New conversations use Astra automatically without requiring a settings change. Both agent paths use medium reasoning for Astra. Existing GPT models remain available, and existing model identifiers keep their previous wire values. (#566)
Claude Fable 5.1 also replaces Fable 5. Existing selections of Fable 5 continue working and now route to Fable 5.1 without requiring the user to select another model. (#468)
Hyper now connects to WhatsApp, Telegram, LangSmith, Datadog, Resend, Airtable, Todoist, and Cal.com.
Hyper can now connect to a personal WhatsApp account through a hosted QR or pairing-code flow.
The connector synchronizes the account’s existing chats and messages rather than using a separate business number. Hyper can search conversations, read messages and participants, and use supported messaging actions with approval.
WhatsApp connections and indexed conversations remain private. Reads are deliberately paced to reduce the risk of the provider restricting the connected phone number. (#440, #485)
Hyper can now connect to a personal Telegram account.
It synchronizes the account’s chats and messages, identifies people using their chosen name, username, or phone number, and supports approved actions through Telegram’s connected-account interface.
Telegram uses the same account-bound execution boundary as WhatsApp. A request prepared for one connector cannot be executed against an account from the other connector. (#447, #485)
Hyper can now connect to a LangSmith workspace through OAuth.
The connector can work with LangSmith runs, thread history, prompts, datasets, examples, experiments, engine issues, and usage information. Sandbox operations remain separately approval-gated because their effect depends on the command or action provided at runtime.
The connection is bound to the workspace selected during authorization, and renewal verifies that the token still belongs to that workspace. (#450)
Hyper can now connect to Datadog and use its remote Model Context Protocol interface.
The connector covers Datadog’s available tools for logs, metrics, monitors, alerts, cases, error tracking, notebooks, incidents, and related operational analysis. The current exposed Datadog tool surface is read-only.
The connection is bound to the organization selected during authorization. (#454)
Hyper can now connect to Resend.
It can work with emails, broadcasts, contacts, audiences, templates, domains, API keys, webhooks, and other resources exposed by Resend’s remote interface.
Read operations can proceed normally. Sending email and changing or deleting Resend resources require per-run approval. The interface is pinned and classified operation by operation so a newly introduced write cannot silently appear as a read. (#466)
Hyper can now connect to Airtable and work with bases, schemas, tables, records, and record comments.
It can inspect base structure, query records with fields, sorting, limits, and formulas, and perform approved record and comment actions. Airtable is an on-demand connector and does not copy bases into Hyper through background synchronization. (#534)
Hyper can now connect to Todoist as a full read-write connector.
It supports tasks, projects, sections, labels, comments, and related task-management operations. Projects and tasks are synchronized into Hyper, while signed webhooks update individual items shortly after they change.
Create, update, complete, and delete actions remain subject to connector approval. (#535)
Hyper can now connect to Cal.com.
Bookings are synchronized into Hyper using incremental updates, and booking-created, rescheduled, and cancelled webhooks refresh the affected booking directly. Hyper can also use Cal.com’s supported API actions with approval.
Cal.com is available in both stable and nightly builds. (#539, #557)
Hyper’s Slack connector can now perform checked message writes such as chat.postMessage.
Every Slack state change requires per-run approval for the exact method, path, and body. Hyper does not treat a general earlier approval as permission for a different Slack write. (#540)
Slack can also search and synchronize private direct messages. Direct messages remain private inside Hyper rather than becoming shared workspace information.
Existing Slack connections may need to reconnect to grant the additional direct-message scopes.
This release completes the move of connector synchronization onto one shared system.
Instead of repeatedly crawling an entire account, supported connectors now use the provider’s change feed, modification watermark, webhook, or item reference to read only what changed.
The new incremental paths include:
Scheduled incremental scans remain the correctness backbone. Webhooks lower the time between a provider change and Hyper seeing it, while periodic reconciliation catches missed deliveries and hard deletions.
Gmail, Google Calendar, Google Drive, Notion, GitHub, Jira, Linear, Slack, Stripe, HubSpot, Attio, LinkedIn, Telegram, and WhatsApp now route deliveries through one connector webhook system.
Each connector declares:
Invalid deliveries are refused without revealing whether a secret, account, or route exists. Repeated events are safe because the provider notification points Hyper back to the current source of truth rather than being trusted as the final stored record. (#461, #465, #470, #471, #475, #476, #477, #486, #504)
Webhook registrations for Google, Attio, Jira, LinkedIn, Telegram, and WhatsApp also use one declared renewal engine. Registrations are renewed under a per-connection lock, with a new registration created before an old one is released. (#516)
Attio webhook signing secrets are now encrypted at rest. Existing plaintext registrations replace themselves with newly issued encrypted secrets during the normal renewal sweep. (#523)
Connector responses now preserve their status, headers, operation, bounded body, and protocol-specific signals until the connector classifies what happened.
The shared synchronization layer then decides whether to:
This prevents a connector from accidentally treating a rate limit or server error as a missing item. (#444)
Model Context Protocol providers can report a rate limit inside an otherwise successful HTTP response. Granola, Krisp, and Wispr Flow now recognize those in-band refusals and back off instead of ending the sync as invalid data. (#520)
Workers now process batches of connector steps during one claim rather than waiting between every provider request. This substantially increases throughput for large Notion, Slack, PostHog, Gmail, and meeting imports while preserving connector-specific pacing.
Rate limits without an explicit retry time use escalating pauses, and server restarts avoid launching unnecessary catch-up work for recently completed connections. (#536, #561)
We also fixed production-specific failures involving Jira’s new bounded-query requirement, incomplete Notion Markdown exports, persisted Notion sync steps, and removals in Krisp and Wispr Flow. (#452, #513, #517, #518)
Every production connector now uses the same authorization, execution, synchronization, webhook, and lease foundations.
The complete connectors-v1 implementation has been removed, including its duplicate sync scheduler, webhook handlers, provider clients, OAuth paths, and hand-built agent tools. Stable now has one connector implementation rather than old and new systems running beside one another. (#500, #533)
Connector code is now separated into:
Each provider has one declaration covering its authentication, interface, execution guide, synchronization, webhooks, leases, and validation. Adding a new provider requires an explicit answer for every capability rather than updating several unrelated registries. (#510, #512)
The Search button and Command-K now search conversation titles and message content across the workspace. Results highlight matching words, show a relevant message excerpt, and open the selected conversation. An empty search shows recent chats. (#439)
Chats can now be added to Favorites from the conversation header or row menu. Favorites synchronize through the server and appear consistently across devices. (#480)
Desktop keyboard shortcuts now support:
The chat core owns navigation history, so sending from the home screen, deleting a conversation, changing workspaces, and opening an in-progress chat keep the navigation stack consistent. (#451, #542, #567)
Files can now be dragged directly into a conversation. Hyper shows an immediate pending attachment while it reads the file, then replaces it with the normal attachment once it is ready. (#455)
Hyper can also read:
.xlsx workbooks, preserving separate sheets and tabular rows..pptx presentations, preserving slide order, slide text, and speaker notes.Both formats use bounded archive extraction to limit entry counts and decompressed size before their contents reach the agent. Legacy .xls and .ppt files remain unsupported. (#448)
A file, pasted passage, or feedback comment can now be sent without typing an additional text message beside it. Hyper uses the attachment or comment itself as the user’s request context. (#449)
Feedback comments now use the same compact chip treatment as file attachments. Writing and reading a comment opens beside the selected passage without covering it, and the highlighted text remains visible while feedback is being written. (#443, #451)
We also refined the sidebar, title bar, navigation controls, chat groups, search controls, shortcuts, themes, loading states, and compact interaction components. (#472, #567, #572)
Application state now lives in the shared Rust core and reaches the interface as revision-stamped state frames. A snapshot restores any update the frontend missed, while session generations prevent abandoned work from a previous account or workspace from replacing the current state. This addresses pages that could previously remain stuck because they missed one pushed event. (#266)
Some older Anthropic conversations contained unsigned reasoning history that newer provider APIs refuse.
Hyper now removes incompatible unsigned reasoning blocks while preserving valid signed, encrypted, and redacted reasoning. If an old assistant message contains no replayable content after repair, Hyper omits that message instead of making the entire conversation unusable. (#436)
We also upgraded Rig from 0.39 to 0.42. Existing stored messages are converted when loaded, strict tool schemas now use Rig’s native support, and old tool calls and results are paired into the shapes the current runtime expects. (#460)
Stable releases now come only from the protected stable branch.
The release system supports automated backports, explicit promotion pull requests, stable-branch CI, release-document validation, and rules preventing database migrations from entering a patch release accidentally. Promotion pull requests now include the complete merge, validation, and deployment procedure. (#437, #577)
Every connector sync now reports structured progress, completion, retry, and terminal failure events after the corresponding database transaction commits. A read-only CLI report combines those events with connection, job, lease, rate-limit, and configuration information to show the live health of each connector. (#499, #541)
Database-backed server and synchronization tests now run in CI against a real Supabase database. Generated TypeScript database types are also rebuilt from the migrations and checked for drift rather than being generated from a developer’s local database. (#555, #559, #568)
The last release made Hyper capable of doing substantially more work during a conversation. This release allows that work to continue after the conversation ends.
Automations are one of the biggest features we have shipped. Hyper can now continue doing real work after you close the app while giving you clear control over its instructions, accounts, actions, permissions, and results.
We are continuing to expand the events that can trigger automations, the work they can complete safely, and the connectors available to every team.
If there is a connector, automation, or workflow your team needs, tell us at founders@heyhyper.ai.
Full changelog: v1.4.0...v1.4.1
v1.4.0
The last release added important connectors, clearer access controls, and more reliable synchronization. It also showed us the next set of problems: Hyper could only use a limited part of many connected products, reviewing a long answer required repeated back-and-forth, multiple accounts were difficult to manage, and several parts of the app still felt unfinished.
Our goal for this release was to address those problems directly:
This release turns on the new connector system we have been building over the past several weeks.
Previously, many connectors exposed a small collection of tools that Hyper’s team had implemented individually. Even when a provider offered hundreds of useful API operations, the agent could only use the handful we had manually added.
The new system works from each provider’s native API definition. Depending on the product, that may be an OpenAPI specification, GraphQL schema, Google Discovery document, Swagger definition, or Model Context Protocol tool catalog.
Hyper compiles those definitions into a bounded interface the agent can search and use. This gives the agent access to a much larger portion of each connected product without requiring us to build every operation as a separate tool. (#325, #336, #363)
The system preserves each provider’s own structure:
Before a request runs, Hyper validates the complete method, path, query, and body against the compiled interface. Requests that fall outside the authorized surface are rejected before credentials are loaded or a network request is made. (#344, #345)
Hyper also tries known requests directly instead of repeatedly searching an interface it already understands. When a provider rejects a request, the useful part of that error reaches the agent so it can correct the request without exposing internal database or infrastructure details. (#357, #372)
This is a major change to what Hyper can do. It also makes new connector capabilities much faster for us to ship: instead of creating an entirely new tool for every action, we can safely expose more of the provider’s existing API through one shared system.
Hyper now connects to HubSpot, LinkedIn, Krisp, Neon, and Wispr Flow.
Hyper can now connect to a HubSpot portal and use its CRM alongside information from the rest of your company.
The connector covers HubSpot’s standard and custom CRM objects, including:
Hyper stores records with the labels, properties, stages, and relationships defined by the connected portal rather than assuming every company uses the same CRM structure.
It can search and read records, inspect schemas and lists, create or update records, add notes, manage tasks, associate records, and archive records when explicitly asked.
HubSpot changes arrive through application webhooks, regular checks for recently modified records, and periodic full reconciliation. This combination covers property changes and custom object types that HubSpot does not expose through its webhook system. (#339, #359)
Hyper can now connect to a LinkedIn account through Unipile.
The connector can work with:
Hyper can search for people, inspect profiles and companies, read conversations, draft or send messages, send connection requests, publish posts, and comment on posts when asked.
The connection runs through a provider-specific security boundary. Hyper binds every request to the LinkedIn account that was connected and does not expose Unipile tenant credentials or account-routing fields to the agent. (#341, #354)
Hyper can now connect to Krisp as a read-only meeting source.
It can:
Owned and shared meetings are indexed into the workspace. Hyper distinguishes between a meeting that does not exist, one that has not been transcribed yet, and one whose transcript was shortened because it exceeded the read limit.
A missing calendar or disabled provider setting is returned as something the user can fix rather than as an unexplained tool failure. (#342, #352, #397)
Hyper can now connect to Neon through Neon’s Model Context Protocol server.
The connector can:
Hyper does not rely only on the permission displayed by Neon’s consent screen. Queries are wrapped as a single read and executed inside a read-only transaction. Results are bounded before they enter the conversation.
Database credentials and connection strings are withheld from the agent. A database can still be addressed directly even when it falls outside a shortened browse result, so listing limits do not become access limits. (#340)
Hyper can now connect to Wispr Flow’s Notetaker and scratchpad.
It synchronizes:
Hyper can search meetings and notes, read a meeting or transcript, list upcoming events, and resolve shared-note links.
Wispr Flow returns long notes and transcripts in character ranges. Hyper follows those ranges with explicit stopping rules, keeps the content in order, and marks a document when the provider stopped before the complete body was available. (#406, #409, #413)
You can now review a Hyper response the way you review a shared document.
Select a passage in an answer, add a comment, and continue reviewing. Each comment appears above the message composer and is sent with your next message.
This makes it possible to review a long draft in one pass:
Each comment keeps the passage it refers to, what you said about it, and its position in the response. Comments are separate from file attachments, with their own count and size limits. (#380, #388, #392)
After a comment is sent, the passage remains highlighted and the comment appears on the message that carried it. Closing and reopening the conversation preserves both the comment and its relationship to the original passage. (#407, #411)
This is especially useful for emails, plans, product documents, investor updates, marketing copy, and other work where several parts of one answer need separate feedback.
A member can now connect more than one account from the same provider.
For example, you can connect:
Each connection gets its own row in Settings. When several connections share a provider, Hyper uses the account name as the row title so they are easy to distinguish. Connect, reconnect, disconnect, and sharing controls apply to the exact row you selected rather than every account from that provider. (#360, #374, #382, #401, #408)
The account is also part of the authorization boundary. Hyper no longer selects an arbitrary connection when several accounts could satisfy a request.
When a request does not make the intended account clear, Hyper lists the available accounts and asks which one to use. It does not silently choose one. This is especially important for actions such as drafting an email, sending a message, or changing a record. (#368, #416, #419, #421, #423, #427)
Reconnects are also tied to the account they started from. If a provider’s consent screen returns a different account, Hyper rejects the renewal rather than updating the wrong connection and reporting success. (#390)
The new connector system distinguishes between reading information and performing an action.
Read-only operations can proceed without an unnecessary prompt. When Hyper wants to perform another kind of operation, it prepares and validates the complete request before asking for permission.
The approval card shows:
You can deny the request, allow it once, or allow the same operation in the future. Durable approvals are tied to the exact connected account and its current authorization. Reconnecting an account invalidates old approvals rather than carrying them onto a new credential. (#338, #343, #345, #346, #348, #351)
Approvals are stored with the conversation, so reopening a previous chat still shows what was requested and whether it was allowed, denied, or expired.
The same approval experience is now available on iPhone as well as desktop. (#378)
Hyper previously displayed every tool call as a separate card with technical fields that were difficult to scan.
Consecutive tool calls now appear as one activity timeline. While Hyper is working, the timeline shows short descriptions and the sources being used. Once the work finishes, it collapses behind a simple “Done working” summary.
Approvals remain visible outside the collapsed timeline because they require a decision from the user.
The timeline also uses the real connector mark for each source, provides clearer progress text, and avoids filling the conversation with raw request and response JSON. (#353)
The iPhone app now uses the same timeline, thinking indicator, provider marks, and approval states as desktop. (#378)
Attachments now keep one consistent appearance before and after they are sent.
Images, PDFs, documents, spreadsheets, and pasted text appear as artifact chips with the file name, type, and an appropriate preview. Sent attachments no longer change into a different generic tile after leaving the composer.
Images open in an image viewer. Text-based files and large pasted passages open in a readable text view. PDFs attached to a draft can be previewed in the app. (#361, #362)
Large pasted passages now become attachments instead of taking over the composer. Smaller pastes remain editable as normal message text. Pasted text also keeps its identity when a conversation is closed and reopened rather than returning as an unnamed file. (#362, #370)
You can now set personal response preferences under Settings → You → Responses.
Preferences include:
Changes apply automatically without a separate save button and sync across desktop sessions.
These preferences affect how Hyper presents its answer. They do not change which tools it uses, what evidence it considers, or which permissions an action requires. A member who leaves everything at the default receives the same behavior as before. (#417)
Hyper is now available on iPhone through TestFlight, at https://testflight.apple.com/join/TCFSyRbj.
The iPhone app uses the same Rust core as the desktop app, giving it the same account, workspace, conversation, and connector state rather than maintaining a separate mobile implementation.
The app supports:
Sign in with Apple uses Apple’s native authorization flow and a nonce-bound identity exchange. Apple’s one-time profile name is saved before the account loads, so private-relay accounts do not have to appear under an unreadable email address. (#324)
Tool progress and approvals now match the desktop experience instead of appearing as incomplete mobile placeholders. (#378)
We also shortened the release path for both apps. Desktop and iPhone builds can now run in parallel, with dedicated caches for Rust and Xcode build output. This reduces the time between a finished change and a TestFlight or desktop build being ready. (#395)
Several controls in Settings previously appeared interactive without saving the result.
Display-name changes now persist through the account’s identity record and appear in workspace rosters after the account reloads. (#389)
Workspace renames now remain visible after server refreshes and cannot be overwritten by an older list request that finishes late. (#399)
You can also upload an account photo. Hyper validates the actual image contents, removes embedded camera metadata, applies orientation, creates a square image, and stores the result under an immutable address. A photo you choose takes precedence over one supplied by the sign-in provider. (#405)
Workspace administrators can upload a workspace image as well. It appears in the sidebar switcher, workspace menu, and Settings. Private workspaces continue using their initials. (#425)
We fixed several issues that made normal use feel rough.
Scrolling upward while Hyper was answering could cause the transcript to pull itself back toward the newest message. Tool timelines opening or closing could also move the text someone was reading.
The transcript now distinguishes between following the latest response and holding the position chosen by the user. While held, it anchors the visible message as content grows or collapses around it.
Keyboard scrolling, small trackpad movements, scrollbar dragging, and the return-to-bottom control now behave consistently. The copy button also remains stable when a streaming response briefly pauses. (#369)
A returning user could briefly see the full sign-in screen while Hyper restored their existing session.
Session restoration now has its own loading state. An account that genuinely cannot load gets a separate screen with Try again and Sign out, rather than disguising recovery as another Google sign-in attempt. (#377)
After installing an update and restarting, Hyper could reopen behind other windows. The restarted app now focuses its main window immediately. (#337)
Each connected account has its own row, account names are used when a provider has several connections, and dialogs identify the exact account being reconnected or removed.
This removes ambiguous rows such as two connections both titled “Gmail” and prevents an operation on one account from making every account from that provider appear busy. (#374, #382, #408)
We rebuilt significant parts of connector synchronization around durable jobs and explicit failure types.
A synchronization failure is now classified as one of several outcomes:
Retries remain durable across server restarts. Webhook and operator requests are also stored before they are acknowledged, and repeated requests are combined instead of creating overlapping scans. (#379)
A shared worker pool now runs indexing across providers instead of relying on separate polling loops. Jobs from the same provider account respect one shared rate authority, while independent accounts can continue in parallel. A rate limit observed through one workspace therefore prevents another workspace from immediately repeating the same request against the same provider account. (#385)
The new indexing system now covers:
Indexing requests are checked against the same compiled provider interfaces used for interactive agent calls. A connector therefore cannot silently drift into making requests that its published contract no longer supports.
Connected products change their API definitions independently of Hyper. A removed field or renamed operation could previously appear only when someone used that connector.
A scheduled connector-drift job now compares the live provider definitions with the immutable versions used by Hyper. It reports all changed or failed connectors in one Slack alert, with bounded diffs in the job summary and complete diffs stored as workflow artifacts. (#364, #365, #366)
Provider definitions are stored as immutable, content-addressed release objects rather than downloaded unpredictably during a release. A release therefore builds against the interface that was reviewed, while the separate drift job tells us when an upstream provider has moved. (#364, #367, #393)
The largest change in Hyper 1.4.0 is not any single connector or interface improvement. It is the new foundation underneath them.
This moves Hyper significantly closer to being an agent that can do real work across your entire company, not only retrieve information from it.
We are continuing to expand safe read and write support across every connector, make connector synchronization increasingly self-healing, and improve the agent’s ability to complete longer and more complicated work.
If there is a connector or workflow your team needs, tell us at founders@heyhyper.ai.
v1.3.0
The last release put the rebuilt Hyper in front of real teams. Those teams quickly showed us where the product still created friction: important tools were missing, connected information could be shared too broadly, synchronization was hard to understand, and direct links did not always open the information Hyper already had.
Our goal for this release was to address those problems directly:
Hyper now connects to Attio, Jira, Confluence, and Supabase.
Hyper can now connect to an Attio workspace and use its customer and relationship data alongside information from the rest of a team’s tools.
The connector indexes:
Hyper can search across Attio, query records with structured filters, read a record together with its related activity, inspect meetings and transcripts, and find email activity involving a person, company, domain, or record.
When explicitly asked, Hyper can also create or update records, add notes and comments, manage tasks, and update list entries such as deal stages. Destructive actions remain limited and explicit. (#260)
Attio changes arrive through signed webhooks, with periodic refreshes for information such as meetings and transcripts that Attio does not publish through webhooks.
Hyper can now connect to Jira sites and work with projects, workflows, issues, custom fields, comments, work logs, attachments, links, boards, and sprints.
It can:
Workflow transitions account for each Jira site’s own fields and statuses instead of assuming every team uses the same process. Hyper refuses ambiguous transitions rather than guessing which workflow action the user intended.
Jira updates arrive through per-site webhooks backed by regular sweeps. This covers missed deliveries, expired registrations, newly added projects, and temporary provider failures. (#317)
Hyper can now connect to Confluence spaces and use pages, blog posts, comments, attachments, and action items.
It can:
Confluence edits are version-checked. Hyper reads the current version before replacing content and refuses to overwrite a newer edit made in the meantime.
Confluence does not provide OAuth applications with the same webhook support as Jira, so Hyper checks for recent changes every 15 minutes and periodically reconciles the full site. (#317)
Hyper can now connect to one or more Supabase projects through OAuth and run read-only SQL through the Supabase Management API.
The connector applies several limits before a query reaches a database:
SELECT statement is accepted.read_only setting.Supabase is an on-demand connector: it does not copy database rows into Hyper or run a background synchronization process. (#290)
A connection no longer has to expose everything it contains to every member of a workspace.
The person who connected an account can now choose between:
The owner always retains access to their own connection. Other members only receive an item when both the connection and that item are classified as shared. This rule is enforced when Hyper searches, reads, resolves links, or accesses connected Google documents, not only in the interface. (#282)
Hyper also preserves privacy signals from each source:
Not every provider exposes item-level permissions. When a provider cannot answer who should see an item, Hyper defaults to the safer interpretation instead of assuming it is shared.
This release is the first version of connector access control. More granular controls by member, role, and information type are still to come.
Users often already know which document, issue, or page they want Hyper to inspect. Searching by title was an unnecessary and unreliable extra step, especially when several files had similar names.
Hyper can now resolve pasted links to information it has already indexed and open the matching source directly.
Resolution supports canonical source URLs across connected services and stable identifiers embedded in links, including:
The lookup remains scoped to the current workspace and applies the same access-control checks as search and reading. A pasted link does not bypass connector privacy. (#277)
Hyper could already create Gmail drafts without sending them. It can now draft a reply to a specific message inside the existing Gmail thread.
The connector reads the source message, preserves the thread, and creates the required reply headers. It rejects malformed source headers and still leaves the message as a draft for the user to review and send. (#272)
We also fixed Gmail synchronization that could leave recent messages unavailable after the initial import. New messages should now continue arriving without requiring a manual reconnect or full synchronization.
Early users could not easily tell whether a connector was actively importing information, had finished, or had simply checked and found nothing new. Notifications were frequent, vague, and disconnected from the rest of the interface.
The sidebar now shows synchronization as a standing activity instead of producing a stream of temporary toasts.
While a connector is actively importing information, Hyper shows a live state such as “Syncing Gmail…”. When synchronization finishes, the sidebar reports which sources changed. Closely timed updates are combined into one message rather than shown individually. (#273, #280)
A completed sync is announced only when it actually stored or removed information. Routine checks that found no changes remain silent.
This required synchronization writes to distinguish between:
That distinction also stops connectors such as PostHog from repeatedly announcing unchanged dashboards after every scheduled poll.
We fixed several connector-specific failures found in production.
PostHog connections that reached more than one organization could previously fail every scheduled sync because the connector expected exactly one organization when naming the connection. Hyper now treats the full authorized organization set as the connection’s identity while continuing to sync every reachable project. Existing affected connections recover automatically. (#275)
We also stabilized the shared scheduler used by polling and webhook-based connectors:
These changes are especially important for Attio, Jira, Confluence, Gmail, Google Drive, and PostHog, which combine different forms of webhooks, polling, refreshes, and full reconciliation.
We made several small changes that remove repeated friction from the desktop app.
The message composer now keeps or restores focus when the app becomes ready, after a message is submitted, and when the user opens or creates a conversation. Users can continue typing without clicking back into the input after normal keyboard-driven navigation. Explicitly clicking elsewhere still moves focus as expected. (#304, #331)
We also made active work easier to distinguish: the stop control now uses a clear destructive treatment instead of looking like a normal primary action. (#273)
We built a native SwiftUI application that uses the same Rust core as the desktop app instead of creating a separate mobile implementation.
The current iPhone app supports:
The shared core owns authentication, workspace state, conversations, and active work. The iPhone app is a native interface over that same state, which reduces the chance that desktop and mobile behave differently. (#263)
The iPhone app is not included in this Mac release. We plan to begin distributing it through TestFlight next week.
The last release showed that teams wanted to use Hyper across more of their actual work, but connector coverage, privacy, and freshness still limited how much they were comfortable connecting.
This release moves each of those constraints forward:
The next step is to deepen what Hyper can safely do inside every connector while making access controls more granular. We are also working on a more sophisticated agent that can be interrupted or redirected while it works and can delegate parts of larger tasks to specialized subagents.
HubSpot, LinkedIn, Resend, and additional connectors are also under consideration. If there is a connector your team needs, tell us at founders@heyhyper.ai.
v1.2.0
The last release gave us the minimum foundation needed to put the rebuilt Hyper in users’ hands. Our goal for this release was to do that: onboard real users, watch where Hyper failed in daily work, and quickly improve the product around them.
We shipped 121 pull requests across five major areas.
Finding the right information is often only the first step. Users still need to turn it into an email, document, task, or product decision.
Hyper can now create and safely update work in the tools teams already use:
Hyper reads the latest version before editing and refuses to overwrite work that changed unexpectedly.
We also added support for attaching Microsoft Word documents to a conversation. (#267)
Getting Hyper to users required more than sending them a download link. Teams needed a complete path from creating a workspace to inviting coworkers and connecting their tools.
New users now create and name a workspace during onboarding. Workspace admins can invite people by email, assign roles, view or cancel pending invitations, and manage existing members. Invitations work for both new and existing Hyper accounts. (#154, #244, #247, #248, #254)
We also fixed a problem that could open an invited user in the wrong workspace and brought account, workspace, member, invitation, and connector settings into one clearer experience. (#259, #262)
Users can now move between conversations or start a new one while Hyper continues working. Each conversation keeps its own draft, progress, response, and cancel control. (#237)
When Hyper uses a previous conversation, it shows that conversation’s name and lets the user open it directly. New conversation titles also appear immediately instead of after several delayed checks. (#203, #206, #213)
The first onboarding sessions exposed several places where connections could appear stuck, finish without updating the interface, or lose progress during a large import.
Connector status now updates immediately, shows clearer states and errors, and cannot be replaced by an older response that arrives late. (#208, #225, #228, #266)
We also made large Gmail imports resumable, added Google Shared Drive synchronization, and made Granola imports less likely to duplicate or lose meetings after an interruption. (#155, #212, #215)
We added analytics for the rebuilt app so we can measure the path from opening Hyper to signing in, completing onboarding, connecting a service, and sending a message. Website activity can also be connected to the same user after sign-in. (#224)
Tracking began on August 13. In the first two days:
Those numbers include internal and test activity, but the usage has already uncovered important problems. Users found cases where Hyper could not open a connected Google Drive file from its URL, missed new Gmail messages, or presented an older source too confidently when newer information existed.
We also made releases faster to build and added automatic Slack notifications when a deployment succeeds or fails. This helps us respond to feedback and ship fixes more regularly. (#200, #220)
In the last release, we wrote:
The next step is to onboard users, watch where the product fails in daily work, and improve it through regular stable releases.
We accomplished the core of that goal. People are now onboarding, connecting their tools, sending messages, and giving us specific feedback from real work. Hyper can also carry more of that work through to completion instead of stopping at an answer.
The product is still early. Our next step is to continue onboarding teams while improving the parts of the experience that now create the most friction:
Full changelog: v1.1.0...v1.1.1
Full changelog: v1.0.1...v1.1.0
Full changelog: v1.0.0...v1.0.1
Full changelog: v0.1.0...v1.0.0
v0.1.0
Our goal this week was to finish the minimum set of features needed to put Hyper in users' hands, then stabilize the infrastructure underneath them.
We focused on four things:
We merged 39 pull requests across those areas this week.
Connected services can now update Hyper automatically instead of waiting for a manual or full import.
We also finished the infrastructure needed to test this behavior safely:
Hyper can now search and read connected GitHub repositories.
It can list the repositories available to a workspace, search their contents, and inspect the relevant sections of source files. This helps Hyper answer questions that require both company discussions and the implementation behind them. (#100)
We also installed the required code-search tools on our servers so this works without any setup from the user. (#106)
Hyper's GitHub access remains read-only.
We improved both the substance of Hyper's answers and the way its work appears in the app.
We strengthened the parts of the app that keep each person's work separate.
Most desktop business logic now runs in Rust rather than inside the interface. Account and workspace changes create isolated sessions for chat, connectors, and drafts. When someone signs out or switches workspaces, Hyper closes the previous session, saves drafts, stops active work, and clears its state from the interface.
We made this change to reduce bugs where work from one account or workspace could remain visible after switching to another. It also gives us a more reliable base for the desktop app. (#88)
Additional account and workspace work includes:
Until now, we only published automatic nightly builds. This week we built the machinery required to publish deliberate, tested stable releases.
We prepared the public website for the new release:
The remaining work reduces deployment risk and makes it faster to find problems before they reach users.
Hyper now has the minimum foundation we need to begin putting the rebuilt product in front of users:
The next step is to onboard users, watch where the product fails in daily work, and improve it through regular stable releases.