Hyper 1.13.0

Goal for this release

The last release brought our new branding across the desktop app and website, paid plans, four new connectors, and faster conversations.

Since that release was only a few days ago, this is a smaller one with two new connectors and improvements to long conversations, model selection, and the app’s new look.

Our goals were to:

• Expand the services Hyper can work with.
• Make long conversations more efficient to resume.
• Remember your model choices as you move between conversations.
• Give our plans new names.
• Refine the new design across the app.
• Improve trial access and long-running connection reliability.

Two new connectors

Hyper now connects to Apollo and Upwork.

Both connectors read information on demand rather than keeping a background search index. Supported actions depend on your connected account’s access and Hyper’s approval controls.

Apollo

Connect Apollo to find and enrich people and companies and work with your sales outreach.

Hyper can read:

• People and company search results.
• Saved contacts, accounts, and lists.
• Deals, tasks, and phone-call records.
• Sequences and their emails.
• Recorded conversations and transcripts.
• Website visitors and sales analytics.
• Connected mailboxes, schedules, and credit usage.

Supported actions include:

• Enriching people and companies with contact and company information.
• Creating and updating contacts, accounts, lists, deals, and tasks.
• Creating and updating sequences.
• Adding contacts to sequences or removing them.
• Starting sequences.
• Drafting and sending email.
• Running Apollo’s AI tasks for prospecting and research.
• Managing supported website visitor tracking and mailbox purchases.

For example:

Find companies that match our ideal customer profile and organize the results into a list.

Review our active sequences and summarize which ones are getting replies.

Enrichment and contact reveals can spend Apollo credits. Starting a sequence sends real email on its schedule, and buying a mailbox incurs a charge.

Apollo connections are private by default. Hyper reads the connected member’s Apollo team on demand. ( #1671)

Upwork

Connect Upwork to work across your client, freelancer, and agency accounts.

Hyper can read:

• Jobs, freelancers, and saved jobs.
• Job posts, proposals, invitations, and offers.
• Contracts, milestones, timesheets, and work diaries.
• Messages and meeting recaps or transcripts.
• Profiles, agencies, teams, and company information.
• Financial summaries, transactions, balances, and Connects.

Supported actions include:

• Creating, updating, publishing, renewing, and closing job posts.
• Inviting freelancers and reviewing or shortlisting proposals.
• Writing and sending proposals.
• Drafting offers.
• Creating, updating, and submitting milestones.
• Pausing, restarting, and ending contracts.
• Sending and editing messages.
• Scheduling, moving, and canceling meetings.
• Updating supported profile, agency, company, and team information.

For example:

Find freelancers who could help us redesign our landing page and prepare a shortlist.

Review my active contracts and summarize the milestones that need attention.

Many changes first produce an Upwork preview for review. Some actions finish on upwork.com, including sending or accepting offers, funding or approving milestones, changing hourly limits, and starting profile boosts.

Submitting or boosting a proposal can spend Connects, and an AI interview uses one of your attempts. Hyper asks before each of these actions every time, even if you chose to always allow that kind of action. Published jobs and sent messages reach other people.

Upwork connections are private by default. Hyper reads Upwork on demand rather than maintaining a synced copy. ( #1677)

More efficient long conversations

Summarize large conversations while they are idle

Hyper can now summarize very large conversations shortly before the model provider’s cached context expires.

This reduces the amount of conversation context that a later turn needs to process.

The automatic summary applies to conversations with at least 250,000 context tokens. It is scheduled after approximately:

• 55 minutes of inactivity for Claude conversations.
• 25 minutes of inactivity for GPT conversations.

Hyper checks that the conversation is idle before scheduling the summary and checks again before accepting it. A new message takes priority.

The summary uses the model that wrote the last reply and follows the same billing checks as other summarization work. If the cache window has already passed, Hyper skips that idle-summary opportunity. ( #1658)

Keep summarized conversations in place

Summarization no longer moves a conversation to the top of the sidebar or marks it unread.

This applies to both automatic idle summaries and manually requested summaries. Your visible conversation history remains available. ( #1653, #1658)

Reuse Claude context for longer

Claude requests now use a one-hour prompt cache instead of a five-minute cache.

This allows more conversations resumed after a short break to reuse cached context rather than processing the same context as a fresh cache write.

Usage accounting has also been updated to reflect the one-hour cache’s write rate. The amount saved depends on the conversation and when you return to it. ( #1645)

Model choices that stick

Remember a separate choice for GPT and Claude

Hyper now remembers your last selected model from each provider.

Previously, choosing a GPT model and then returning to a Claude conversation could cause that conversation to fall back to the first Claude model instead of your previous choice.

Now:

• A new conversation starts with your most recently selected model.
• An existing GPT conversation uses your last selected GPT model.
• An existing Claude conversation uses your last selected Claude model.

These choices persist between app sessions. Existing saved model preferences continue to load. ( #1649)

Restore model choices before applying plan changes

Hyper loads saved model choices before checking them against your current plan.

This prevents startup plan updates from overwriting a saved preference before it has been restored.

When a plan no longer includes a selected model, Hyper still moves that selection to an available model from the same provider. ( #1649)

New plan names

Our plans are now Free, Snap, Crackle, and Pop. The trial is now the Crackle Trial.

The new names appear in the app, on the website’s pricing and billing pages, and on Stripe Checkout and invoices.

Only the names change. Each plan keeps its price, usage, models, and features, and existing subscriptions continue as they are. ( #1673)

More polish across the app

A more consistent icon set

We expanded the comic-style icons across the desktop app.

The update covers:

• Navigation and conversation controls.
• Tool activity and approval controls.
• Hypercards and attachments.
• Settings, billing, and connected accounts.
• Search, feedback, and notifications.
• Rich-text editing controls.

We also refined shared shapes such as arrows, checks, pencils, and chevrons so they look consistent wherever they appear.

Icons with accessibility labels now remain available to assistive technology. ( #1653, #1654)

Refined sidebar and settings navigation

Sidebar controls now use more consistent icons, spacing, and interaction styles.

Working indicators align more cleanly with conversation rows and stay clear of row actions.

Settings navigation buttons remain in place when pressed, matching the behavior of conversation rows. ( #1653, #1654, #1669)

Consistent buttons and control heights

Buttons now share a consistent corner shape across sizes and styles.

Header controls use a common height, and chat-bar controls align with the model picker.

These changes apply across shared controls, including icon buttons and buttons drawn directly within individual screens. ( #1672, #1678, #1679)

Full-screen settings and search backgrounds

The dotted-paper background now extends across the full settings and search screens.

This makes those screens more consistent with the rest of the refreshed app. ( #1664)

Reliability improvements

Trial access updates without a restart

Starting a trial now notifies connected apps to refresh their plan and feature access.

You no longer need to restart the app for a newly granted trial to become available.

Repeated requests to start the same trial preserve its existing dates rather than granting it again. ( #1660)

Fewer interruptions during long-running work

We changed the server connection configuration to prevent the default one-hour connection lifetime from cutting off an active chat stream.

This addresses interruptions caused by the connection aging out while Hyper is still working. ( #1646)

Cleaner attachment lists

Large tool results stored internally for Hyper to inspect no longer appear as downloadable attachments simply because they were saved as files.

Files produced for you remain available through the normal attachment and download flow. ( #1643)

These three fixes already shipped in v1.12.1. They are included here for anyone upgrading from v1.12.0.

Website improvements

A new scene behind each plan

The pricing page now shows an animated Pop Art scene behind each plan, printed in halftone dots and hard ink: a flock of birds for Free, a comic-style blow for Snap, lightning for Crackle, and popping bubbles for Pop.

Choosing a plan brings in its scene, and each plan plays a short sound when you press it. A click anywhere on the pricing page bursts Snap!, Crackle!, and Pop! in turn.

With reduced motion turned on, each plan shows a still panel. Without graphics acceleration, the page shows the plan’s flat color. ( #1676)

Easier-to-read feature descriptions

The descriptions on the website’s “Known powers” screens now stay still instead of scrolling automatically.

We also adjusted the text sizing for different screen sizes. ( #1655)

More reliable expanded screens

Opening a “Known powers” screen now holds the page still behind it.

Expanded screens and their background shading appear above adjacent sections and navigation controls instead of being clipped or covered.

The page also preserves its width when the scrollbar disappears. ( #1663)

Smoother navigation menus

We fixed the pointer gap between the website’s navigation keys and their menus.

You can move the pointer into an open menu without it closing as you cross the gap. ( #1666)

Updated artwork and link previews

We refined the website’s social-preview artwork and added video metadata for services that support animated previews.

Page titles and descriptions now use “The powerful agent for work.”

The marketer illustration has also been updated to match the latest artwork. ( #1648, #1652, #1657, #1659, #1662)

Other additions and improvements

• Trial reminder emails: Members on a trial now get an email 24 hours before it ends and another when it ends, with a link to choose a plan. Members already on a paid plan get neither, and no email is sent for a trial that ended before this release. ( #1670)
• Replies go to both founders: Replies to Hyper’s campaign emails now go to founders@heyhyper.ai, so both founders receive them. ( #1650)
• Clearer connector guidance: Updated the agent’s instructions and product documentation for supported connector operations, account access, approvals, rate limits, and service-reported restrictions. ( #1681)
• Updated connector artwork: Added Apollo and Upwork branding to the desktop interface and iPhone provider assets. ( #1671, #1677)

Engineering and release improvements

New connector authentication support

The connector framework now supports account identification through multiple tool calls within an MCP session.

It also supports OAuth callbacks that must return to a local address. The app receives the callback and forwards it to Hyper’s server to complete the connection.

These changes support Upwork’s sign-in requirements. ( #1677)

Explicit operation classifications for new connectors

Apollo and Upwork tools now have pinned interfaces and explicit operation classifications.

This distinguishes reads, changes, and operations that can spend credits or carry out broader actions.

The integrations include tests for authentication, request handling, and operation classification, along with live API checks. ( #1671, #1677)

More consistent interface checks

We expanded checks for shared visual rules, including button corners, control sizing, settings backgrounds, and icon accessibility.

The shared button component also has a text style that aligns with surrounding prose. ( #1654, #1664, #1678, #1679, #1682)

Expanded product-demo tooling

We improved the tooling used to script, record, and assemble product demonstrations.

The launch-demo scenes now cover concept development, research, scripting, video work, and booking, with browser views shown alongside the conversation.

These are scripted demonstration assets rather than additional product capabilities. ( #1545, #1546)

Have a feature your team needs? Tell us at founders@heyhyper.ai.