Latest Runner updates

When someone replies to a message Runner sent, Runner now notices and picks the conversation back up on its own — across email and chat, including Gmail, Outlook, Slack, WhatsApp, and Telegram. You no longer have to come back and nudge it to keep a back-and-forth going, and the session header shows an eye whenever Runner is watching for a reply.

What changed:

• Replies: Runner watches for replies across Gmail, Outlook, Slack, WhatsApp, and Telegram and automatically resumes the task the moment one arrives, with an eye indicator in the session header so you can see when it's watching
• Tasks: A task can now pause to ask you a question and wait for your answer before continuing, and you can schedule more than one follow-up for the same task
• Sidebar: A new toggle lets you collapse or expand the sidebar, and it stays the way you leave it
• Browser: When a task is waiting in the browser, you now see a screenshot of what it's looking at, in a more compact preview
• Action cards: Cleaner draft previews, and Runner can clear a card on its own once it's no longer needed
• Trials: The trial-end screen recommends a plan based on how much you've actually used, with a direct link to upgrade
• Fixed sessions that could get stuck and stop responding; these now recover on their own
• Fixed task answers sometimes being lost when sent before the task had fully loaded
• General performance and reliability improvements

Why it matters: A reply should keep the conversation moving on its own — you shouldn't have to come back and restart it. Runner now handles the waiting, across every channel you work in.

When Runner drafts an email reply, it now keeps everyone on the thread by default — and the reply card shows you exactly who it's going to, so you can review or trim the recipients before anything sends.

What changed:

• Email: Replies now default to the whole thread, and the reply card shows recipients up front so you can adjust who's included before sending
• Conversations: Conversations now have far more room before Runner summarizes earlier messages, so longer chats stay coherent
• Browser: When a task pauses for you to sign in or solve a captcha, you can open the right Chrome tab in one click, with clearer wording about what to do
• Workflows: Suggested workflows now reliably appear, each with its own icon
• Reporting: Report a problem straight from a task; the report window closes the moment you hit send
• Fixed the email review card sometimes taking up to a minute to appear when sending through Gmail
• Connected accounts that lose access in the background now prompt you to reconnect instead of looking healthy
• General stability and reliability improvements

Why it matters: Replying to a thread should keep the right people in the loop without extra clicks, and the everyday flows — email, longer conversations, and signing in through Chrome — should just work.

Documents Runner creates in Google Docs now keep proper formatting — headings, lists, and emphasis — instead of arriving as flat text. And in long conversations, Runner now holds onto much more of what you've discussed before it needs to condense earlier messages.

What changed:

• Google Docs: Documents Runner drafts now arrive with real formatting — headings, lists, and emphasis — instead of plain text
• Memory: Long conversations now keep far more history before Runner condenses earlier messages
• Fixed creating a Google Calendar event mis-checking for conflicts across time zones
• General stability and reliability improvements

Why it matters: The work Runner produces for you should arrive ready to use — properly formatted documents, and longer conversations that don't lose the thread of what you've discussed.

Runner can now do more on your behalf — placing phone calls, creating Google Calendar events, and finding WhatsApp groups by name — each with a clear card showing what it's about to do. The rest of this release makes browsing steadier and cleans up rough edges across connections and messaging.

What changed:

• Phone calls: Runner can place calls for you, with a clear card showing what it's about to do
• Calendar: Creating a Google Calendar event now shows a confirmation card before it's added
• WhatsApp: Find and message groups by name, not just exact matches
• Answers: Mathematical notation now renders properly in responses instead of showing raw symbols
• Messages: When a contact isn't verified, Runner shows the name it found and tells you when it needs Contacts access
• Fixed browsing getting stuck or failing to recover when the browser was in a bad state
• Fixed connection errors being mislabeled as "missing permissions" when the real cause was different
• General stability and reliability improvements

Why it matters: Runner keeps expanding what it can do for you — calls, calendar events, group messaging — while making each action clear before it happens, so you stay in control of what gets done on your behalf.

If you use a Runner-managed connection, Runner now picks the best model for each task automatically instead of staying locked to an older default — so you get the right balance of speed and capability without thinking about it. The rest of this release is label, settings, and reliability polish.

What changed:

• Models: Managed connections now default to automatic model selection, letting Runner choose the best model for each task
• Labels: "Create label" is now available directly from the label menu
• Fixed the workspace default folder picker not working correctly
• Made the empty-session help tip quieter and less distracting
• General stability and reliability improvements

Why it matters: Picking the right model is one less thing to think about — Runner now balances speed and capability for you on managed connections, so you can focus on the work instead of the settings.

The screens you see when getting started with Runner are cleaner and more responsive — the welcome screen's referral card and the start carousel navigation have been polished, and taps near the edges now register reliably. The rest of this release is under-the-hood reliability work.

What changed:

• Getting started: A more polished welcome screen referral card and clearer carousel navigation on the start screen
• Fixed taps near the edge of the start screen not always registering
• General stability and reliability improvements

Why it matters: First impressions count — a cleaner, more responsive welcome and start experience makes getting going with Runner feel effortless.

When Runner drafts a reply into a Slack thread, the preview now shows the whole conversation — the original message on top and every reply beneath it — so you can approve the reply with full context. Getting started is also one step shorter, and a fix unblocks older iMessage drafts that refused to send.

What changed:

• Slack: Replies into a thread now show the full thread in the preview — the original message plus its replies, opened at the newest one — instead of a flat snippet of the channel
• Getting started: Setup is one step shorter — we no longer ask how you heard about Runner
• Referrals: Invite codes are now entered on the web right after you sign up, instead of during desktop setup
• Fixed iMessage drafts created on older versions failing with "This card is missing details needed to send"
• General stability and reliability improvements

Why it matters: Approving a reply is only safe when you can see what you're replying to — the full thread in the preview means no more guessing what came before.

Runner now keeps your conversations steadier — generated images stay put instead of flickering while you type, and long messages show in full. We also fixed iPhone photo sharing and a handful of connection and display issues, so day-to-day work feels more reliable.

What changed:

• Connections: Connectors can now run code to filter and transform large results, so big data sets don't flood the conversation
• Settings: Clearer Mobile Companion settings explaining how it works and which apps it supports
• Conversations: A helpful tip now points you to the Help button when you start a new conversation
• Fixed iPhone (HEIC) photos being silently dropped instead of sent to the assistant
• Fixed generated images flickering or reloading while you type
• Fixed long messages and conversation titles getting cut off
• Fixed some connections, like Figma, appearing broken right after you add them
• General stability and reliability improvements

Why it matters: The things you look at most — your conversation, your images, your messages — now stay put and show in full, so everyday work feels more reliable.

Runner now picks the right model for each request automatically, so you get the best results without choosing one yourself. You can also attach files directly to Slack messages, and sharing files from a second drive or network share now just works.

What changed:

• Model selection: Chat now defaults to Auto, choosing the best model for each request instead of a fixed one
• Slack: Attach files to your Slack messages directly from a message card
• Connected apps: Recipients now show real names instead of internal IDs, with consistent labeling across your messaging apps
• Fixed sending to a Slack channel by name sometimes failing instead of posting to the channel
• Fixed attaching a file stored outside your home folder — on a second drive, network share, or relocated OneDrive — failing to upload
• General stability and reliability improvements

Why it matters: Smarter defaults mean Runner does the right thing without extra choices, and smoother sharing means the files and messages you send just go through.

This release is all about reliability. Runner now works more smoothly with Outlook and loads your message history more dependably, so the apps you connect just feel steadier day to day.

What changed:

• Outlook: A more focused, reliable connection means faster, steadier results when Runner works with your email
• Fixed an issue where some message threads could stop short instead of loading the full conversation
• General stability and reliability improvements

Why it matters: Reliable connections mean Runner does what you ask without surprises — the apps you depend on stay steady and your message history loads in full.

Runner now gives your Google results distinct icons — so Gmail, Calendar, and Drive are easy to tell apart the moment they appear. This release also tightens reliability across your conversations, settings, and file attachments.

What changed:

• Work results: Distinct icons for Gmail, Calendar, and Drive, so you can tell them apart at a glance
• Settings: A clearer, more secure view of your connection key status
• Conversations: Titles for task-based conversations now stay in sync
• Fixed an issue where an attachment that failed to upload could go unnoticed
• Fixed an issue where generating an image could occasionally time out
• General stability improvements

Why it matters: Distinct result icons make it easier to scan what Runner found across your connected apps, while clearer settings and steadier reliability help you trust what Runner is doing.

Runner now offers Claude Opus 4.8 — Anthropic's most capable model — so you can reach for extra reasoning power whenever a task calls for it. This release also makes action cards simpler to scan and approve, and keeps your conversations clean and easy to follow.

What changed:

• Models: Choose Claude Opus 4.8 for your most demanding tasks
• Action cards: Cleaner, simpler cards that are easier to scan and approve at a glance
• Action cards: Sending to a WhatsApp group now shows the group name on the card, so you can confirm exactly where your message is going
• Settings: Simpler, clearer browser profile controls
• Conversations: A tidier chat — internal routing labels no longer appear while Runner is replying
• Fixed an issue where a generated file, like a PDF, could fail to open from a conversation
• Fixed an issue where a saved automation could fail to run because of an outdated model selection
• Fixed the placement of the back button on the connect screen during setup
• General stability improvements

Why it matters: The right model for the moment makes Runner more capable on hard problems, while clearer action cards and a tidier chat make it easier to see what Runner is doing and approve it with confidence.

Runner now greets new workspaces with a guided first run — it helps you pick a skill, connect the right accounts, and see Runner working for you from your very first conversation. This release also tightens reliability across your conversations and browser profiles.

What changed:

• Getting started: A guided first run that helps you pick a skill, connect the right accounts, and see Runner in action right away
• Fixed an issue where a branched conversation could disappear while Runner was replying
• Browser profiles now stay correctly scoped to your active workspace, with a cleaner profile picker in Settings
• Smoother task detail view when collapsing on the home screen
• General stability improvements, including better handling of brief connection hiccups

Why it matters: Runner is most useful when its value is clear from the very first conversation. A guided first run helps you start with the right skill and the right accounts connected — so Runner is working for you right away.

Runner can now send attachments directly in your conversations on WhatsApp, Telegram, Instagram, and LinkedIn — share a document, image, or file without leaving the chat. This release also polishes the daily brief, adds a browser profile picker in Settings, brings more model connections, and tightens reliability across replies, updates, and conversation recovery.

What changed:

• Messaging: Send attachments — documents, images, and files — in your WhatsApp, Telegram, Instagram, and LinkedIn conversations
• Daily brief: A cleaner, clearer brief that's easier to read at a glance
• Settings: Choose which browser profile Runner uses when it browses on your behalf
• Connections: More model options, including Morph and MiniMax
• Fixed an issue where replies in connected apps could lose the original message's context
• Fixed WhatsApp conversations that failed to open for some phone numbers
• Fixed long option labels so their full text now shows on hover
• Fixed updates so they apply reliably instead of occasionally keeping an older version
• General stability improvements, including recovering conversations that previously got stuck and smoother handling of interruptions

Why it matters: Runner is most useful when it can act where your work actually happens. Sending attachments across your messaging apps means you can hand off a file in the same conversation you're already having — no switching apps, no leaving the chat.

You can now attach files to the emails Runner sends — drop in a PDF, an image, or a document and it goes out with the message. Conversations are more flexible too: branch from any earlier message to explore a different direction without losing your original thread.

What changed:

• Email: Attach files to Gmail and Outlook messages before they send
• Conversations: Branch from any message to try a new direction while keeping the original
• Models: Set a default model so new conversations start the way you prefer
• Sign in: A smoother, more reliable sign-in and recovery flow
• iMessage: Runner picks up the right chat context automatically and confirms each message was delivered
• Fixed a sent message being reported as failed even though it went through
• Fixed email attachments that could be lost when reopening a draft
• Restored conversation details that could go missing after a reload

Why it matters: Real work comes with attachments and second thoughts. Sending files with your emails means you don't have to leave Runner to finish the job, and branching lets you explore an alternative without throwing away the thread you started.

Long conversations no longer get cut short, and the assistant's step-by-step thinking is tucked away by default — so the chat stays focused on what matters. And when you send something, you can see its status the moment you hit send.

What changed:

• Chat: Longer conversations no longer get cut short, and the assistant's thinking is tucked away by default for a cleaner read
• Sends: Action cards now show send status the moment you hit send, and iMessage recipients are checked against your contacts before sending
• Automations: New automations no longer pick up unwanted default labels
• Fixed conversations that could get stuck mid-task and stop responding
• Fixed slow or missing suggestions when opening the assistant menu
• Fixed onboarding showing the wrong state for people who'd already connected
• Files created during a conversation now stay reliably linked, even if an import hits a snag
• Scheduled runs now clearly tell you when they can't run because Full Disk Access is turned off, instead of failing quietly

Why it matters: A long working session shouldn't hit an invisible ceiling or drown you in the assistant's scratch work. This release keeps conversations going as long as you need and the chat focused on the answer — while making sure what you send reports its status right away.

The assistant can now read from Instagram and TikTok, giving you more places to pull context from when you're researching or following up. This release also tightens the messaging path across LinkedIn, WhatsApp, iMessage, Instagram, and Telegram — recipients are validated when a card is created instead of failing silently at send time, action cards send with less perceived lag, and revoked sign-ins now lead cleanly to Reconnect.

What changed:

• Sources: Read tools for Instagram and TikTok, so the assistant can pull posts and profiles into context
• Conversations: Default handoff sessions start automatically, and continuing a chat now stays on the default connection instead of prompting
• Messaging: WhatsApp, Telegram, and Instagram action cards reject malformed recipients up front, so a card can't be sent to a chat ID the platform won't accept
• LinkedIn: Action cards refuse non-1st-degree DMs at the moment they're created, with a clearer reason instead of a generic send failure
• iMessage: Send classification is sturdier on the action card path, reducing cases where the wrong number gets picked
• Email: Reply intent is now structural in Gmail and Outlook sends, so replies thread to the right message
• Action cards: Send feels faster — the UI advances optimistically while the request completes, and errors show up as a clearer chip plus a restyled toast
• Action cards: After a card executes, the assistant follows up on what it just did so the next step doesn't get lost
• Connections: Sign-ins flagged as revoked now route directly to Reconnect instead of looking like an outage, and stuck rows heal automatically
• Fixed a billing miscount on image attachments that could trigger quota errors during normal use
• Fixed WhatsApp group messages that were unreadable so they render as honest "unreadable" markers instead of empty content
• Fixed custom tools failing to resolve the right user in workspaces with multiple members
• Fixed idle session failures so the error message reflects what actually went wrong instead of a generic timeout

Why it matters: Runner is most useful when it can reach the places your work actually happens. Reading from Instagram and TikTok adds two more sources of context, and validating recipients the moment a card is created means a bad address gets caught while you can still fix it — not after the send quietly fails.

A small release focused on reliability. When an action card fails to send, the toast now says "Send failed" instead of contradicting itself with a green "Message sent" confirmation. The composer's goal status rail is also back to its compact look, with a live elapsed-time label and a one-click way to clear the active goal.

What changed:

• Sends: Action cards that failed to send no longer flash a green "Message sent" toast — failed sends now show a "Send failed" toast with the underlying reason, and the card stays open with the failure detail
• Goals: Restored the composer's compact goal status rail, including the live elapsed-time label and the one-click clear-goal action when a goal is active
• Fixed more transcript-drift cases so conversations recover cleanly when a session resumes after a brief disconnect instead of getting stuck

Why it matters: A green checkmark on a message that never sent is worse than no signal at all. This release makes send status tell the truth — failures look like failures, and the card stays put so you can act — so you can trust what Runner says happened.

This release is mostly about reliability. LinkedIn sends no longer fail with confusing "invalid recipient" errors, sources whose sign-in has been revoked correctly offer a Reconnect button instead of "Outage", account labels for Attio, Notion, and other connected apps stay accurate, and conversations with Haiku compact at the right time so you don't hit unexpected context-limit errors. You can also now start a new session directly from a label group in the sidebar.

What changed:

• Sidebar: Start a new session directly from any label group
• Connections: A new Claap starter
• Connections: The assistant now reliably recognizes the apps you have connected — asking it to "check my Slack" or "look in my Calendar" routes correctly the first time
• Connections: When a local connection can't start, the message is gentler and Settings only offers a repair flow when there's something you can actually fix
• Conversations: Conversations with Haiku now compact at the right time so you don't hit unexpected context-limit errors, and unsupported window sizes are hidden in the model dropdown
• Fixed account labels for Attio, Notion, and other connected apps so the name you see matches the account you signed in with
• Fixed first-party connections so account labels stay accurate after reconnect
• Fixed LinkedIn sends that failed with an "invalid recipient" error, including a retry loop that could keep trying after the assistant had given up
• Fixed sources whose sign-in had been revoked so they show a Reconnect button instead of "Outage" — and clicking Reconnect now completes cleanly
• Fixed Cmd+Enter on action cards so it advances to the next queued card after a successful send, and keeps the current card open if a send fails
• Fixed transcripts that could drift when only part of a turn made it back from the assistant
• Fixed the assistant pausing unexpectedly while the browser was loading
• Fixed image attachments being billed at their text-equivalent size, which could trigger unexpected quota errors
• Fixed label rename and reorder controls in settings
• Fixed error messages for prompts that exceed the model's context window so they no longer suggest unrelated fixes

Why it matters: Reliability is what makes Runner trustworthy enough to hand work to. This release fixes the small lies — a green checkmark on a failed send, an "Outage" that's really a revoked login, a label that names the wrong account — that quietly erode that trust.

Connecting your apps just got faster. Google Calendar, Google Workspace, GitHub, HubSpot, Salesforce, and Slack now use the same managed connections layer, so they sign in once and stay reliably connected. There are also new starters for WHOOP and Reddit, a growing library of community-built connections, and clearer guidance the first time you set up a connection that needs credentials.

What changed:

• Connections: Calendar, Workspace, GitHub, HubSpot, Salesforce, and Slack now connect through the managed connections layer for faster setup and steadier sign-ins
• Connections: New starters for WHOOP and a curated set of community-built connections
• Connections: When a connection needs credentials, Runner now walks you through setup with a dedicated help conversation
• Local connections: Search across all available starters when adding a local connection
• Local connections: A broken local connection no longer blocks the rest of your session from starting
• Tools: Reddit tools are now available for the assistant to use
• Photos: Better handling of photos shared into Runner, including improved previews and downloads
• Privacy: Draft fields in action cards are now encrypted at rest
• Fixed action cards that could overlap the chat panel header
• Fixed action card "From" labels and improved the error shown when a connection is disconnected mid-send
• Fixed Gmail messages whose text bodies contained raw HTML — Runner now shows clean text
• Fixed Twitter drafts to reject invalid participant identifiers before sending
• Fixed local connection settings so only one row stays expanded at a time, and removed a redundant cancel button

Why it matters: A connection you have to keep re-authorizing isn't really connected. Moving the apps you use most onto the managed layer means they sign in once and stay that way — and when a setup does need your credentials, Runner now walks you through it instead of leaving you to guess.

Your task home now opens to a tidy list, splits In Progress into clear substates, and groups automation runs under the automation they came from. Background tasks now show their tool activity and lifecycle in the transcript, with controls to pause and resume them.

What changed:

• Tasks: Home opens to a list by default, with In Progress split into clear substates
• Automations: Runs from the same automation collapse under a single entry in the sidebar
• Background tasks: Tool activity and lifecycle from background tasks now show in the transcript, with controls to manage them
• Todos: Edit suggested todo titles before accepting them; scheduled dispatches lean toward earlier check-ins
• Tools: Edit local tool servers from settings, with visibility into which tools are blocked by the edit policy
• Compaction: After repeated compactions, the window tightens on its own; the usage tip stays centered when it repeats
• Context: The 1M context window only turns on for models that support it
• Heartbeat: The sweep summary reads as a chief-of-staff briefing rather than a status dump
• Reliability: Scheduled runs wait until the device wakes stably; Runner's own tools come up immediately on launch
• Editing: Markdown edits in chat are no longer capped by the merge policy
• Fixed transcripts that didn't recover after a session moved working directories, or that loaded with parts missing
• Fixed session model selection that ignored the default connection
• Fixed misaligned action buttons in expanded tool settings
• Fixed voice transcription that could fail on certain inputs

Why it matters: As Runner does more in the background, you need to see what it's doing without digging. The redesigned task home and in-transcript background activity make the work legible at a glance — and the clearer substates tell you what's actually moving versus waiting on you.

Every file the assistant produces now lives in a single workspace browser, with CSV tables and video playback built in. Your task list gets a new agent menu that proposes bulk edits — rename, reschedule, regroup — and applies them in one pass after you approve.

What changed:

• Files: A new workspace files browser with inline previews for CSVs and videos
• Tasks: An agent menu suggests bulk task edits and applies them after you approve
• Chat: A per-session context size control lets you tune how much room the assistant has to think on a given conversation
• Compaction: Conversations auto-tighten earlier when you compact repeatedly, with clearer copy in the usage tip and one-tap access to the setting
• Todos: Schedule grouping in the todo view, a tidier model selector, and a composer that matches the session styling
• Todos: Scheduled dispatches default to the current task instead of clearing it
• Heartbeat: Failed runs are retried before new work picks up
• Sign-in: Google sign-in now has a Cancel button so you can back out of the flow cleanly
• Billing: Billing errors in chat now include a direct Billing button to open the portal
• Models: Sonnet with an Opus advisor is now the global default; Opus medium defaults are restored
• Fixed editing a scheduled dispatch from the badge cancelling it instead
• Fixed transcripts that didn't recover correctly from the current session directory
• Fixed referral invite copy and the invitee credit display on the welcome screen
• Fixed the chat compaction tip so the token threshold is clearer
• Fixed the idle watchdog waiting on follow-up questions longer than intended
• Fixed local task controls appearing on sessions that aren't tied to a task

Why it matters: When the assistant makes things for you — reports, exports, clips — they should be one click away, not scattered. The workspace browser puts every output in one place, and bulk task edits let you reshape a whole list in a single approved step instead of one item at a time.

Tasks are now a real place to organize what you're working on, with labels, project assignment, status filtering, retries, and a Today view scoped to your local day. The dashboard gets a new Inbox card surfacing what needs your attention, and the todo list picks up cleaner project grouping and steadier rendering throughout.

What changed:

• Tasks: Tasks now support labels, project assignment, status and source filtering, and a Today view scoped to your local day
• Dashboard: A new Inbox task card surfaces what needs your attention right from the dashboard
• Todos: Cleaner project grouping, unified state indicators, and a polished row design across the todo list
• Heartbeat: Periodic check-ins use Sonnet with an Opus advisor for sharper steering and handoffs
• AI settings: A new default for context size, so the assistant has the right amount of room to think
• Connections: Trusted GitHub setup is now a one-tap flow from the connection catalog
• Sessions: Session label groups remember their collapsed state between visits
• Chat: Retired the legacy Advanced Mode setting in favor of the unified composer
• Fixed switching the default model rebinding older conversations to the wrong setup
• Fixed scheduled dispatches that occasionally didn't refresh their badges
• Fixed the todo session header so it stays visible as you scroll
• Fixed managed conversations that could fail on certain endpoints
• Fixed pending input action icons disappearing while you were typing
• Fixed scrolling on the Labels page in Settings
• Fixed local setups that occasionally created duplicate credential slots
• Fixed the assistant not always collecting answers to its follow-up questions
• Fixed todo agents that sometimes ignored their scheduled routing
• Fixed Mac builds so the notarization ticket is stapled inside the DMG for cleaner installs

Why it matters: As Runner takes on more of your work, you need somewhere to see and steer it. Tasks now organize the way you already think — by project, label, and what's due today — and the dashboard Inbox keeps the next thing in front of you instead of buried in a list.

Adding a new connection is now a single, clean flow with a wider starter gallery and proper credential storage for custom setups. Automations get sharper defaults and per-run working directories, past conversations restore reliably even after switching models, and the assistant explains itself better when something goes wrong.

What changed:

• Connections: A streamlined Add Connection flow with an expanded starter gallery and a generic option for setups not yet in the catalog
• Custom connections: Custom setups now save their credentials, edit their config safely, and recover cleanly from sign-in hiccups
• Automations: A per-automation working directory, a thinking-level control, and smarter defaults for new automations
• Conversations: Past conversations now reload reliably, even after you switch the underlying model
• Heartbeat: Cleaner onboarding states and tighter controls for periodic check-ins
• Settings: Open the Runner Chrome browser from General settings, plus clearer copy on the live-chat row
• Images: New image generations default to GPT Image 2, and large image attachments are resized before sending
• Plans: New Iron, Titan, and Olympus tiers for people who need more capacity
• Fixed the assistant surfacing generic errors instead of the underlying API details
• Fixed automations that occasionally stalled when a handoff failed repeatedly
• Fixed connections that retried with the wrong parameters during a sign-in refresh
• Fixed Google Calendar links so they show the right favicon in finished-work cards
• Fixed an edge case where image attachments larger than the limit would fail silently

Why it matters: Connecting your apps is the first thing most people do, so it should be one clear step instead of a guessing game. These changes make setup faster, keep your custom credentials safe, and make sure the work you've already done — past conversations and running automations — holds up when you switch models or hit a snag.

Runner can now search and read the open web when an answer isn't in your connected apps, and it quietly learns context from those apps so it doesn't keep asking the same questions. Slack drafts also reply in the right thread, and downloads route to the right Mac build.

What changed:

• Web reach: Runner can search the web and read pages when the answer isn't in your connected apps
• Context: Runner learns about you and your work from the apps you've connected, so it doesn't keep re-asking the same questions
• Slack: Replies now stay in the thread you started instead of falling back to the channel
• Connections: Custom integrations now show per workspace, so configuring one workspace doesn't leak into another
• Heartbeat: Clearer controls for pausing or opting out of periodic check-ins
• Help: A new "Talk to a human" card surfaces when you need real support, with a one-click handoff
• Fixed the assistant occasionally losing its tool settings when you switched the default model
• Fixed Slack messages that sometimes silently retried instead of surfacing the failure
• Fixed Mac downloads that occasionally picked the wrong build for Apple Silicon
• Fixed download links that could go stale between releases
• Fixed dark-mode contrast on the Operator phone activation button
• Fixed an edge case where some tool inputs weren't validated correctly before being sent to the assistant

Why it matters: Runner is most useful when it can reach past your connected apps and remember what it has already learned. Web search and carried-over context cut the back-and-forth, and the Slack and download fixes make everyday handoffs land where you expect.