What's new in Supercenter.

Self-checking steps

Agents now verify the result of each step as they work. When a check can't be confirmed, the run pauses and asks a person instead of guessing — you get pulled in only when it actually matters.

A run page you can skim

Runs now name the step that failed and why, with a receipt for every action the agent took. The Command Center gains a unified Outputs tab and a live status summary, so you can see what a run produced without reading the whole transcript.

Improvements

• Scheduled sessions and Slack conversations get their own icons in the sidebar, so it's easier to tell what you're looking at.
• New accounts land in a ready-made workspace with suggested apps for your first daily digest.
• Long-running work recovers from interruptions and resumes where it left off instead of starting over.
• Steps that loop without making progress stop on their own instead of running forever.

Fixes

• The homepage is easier to read on phones — bigger type, tighter visuals.

Automatic improvement, on your terms

Agents can now propose improved versions of their own instructions and working steps. Preview mode lets you review every suggestion before it goes live; On mode applies reviewed improvements automatically; Off keeps things exactly as you set them.

Clearer approvals

Approval requests are now separated from run activity and spelled out in plain language, so it's obvious what you're agreeing to before an agent acts.

Improvements

• Runs blocked by an expired connection say so up front and link straight to reconnecting.
• Set a default thinking level per model, so each agent reasons as deeply as the job needs.
• You can point runs at an optional help destination and resume interrupted requests.
• New connector: Tella.
• The app does noticeably less work behind the scenes — pages load faster and live updates arrive sooner.

Fixes

• Disconnecting an app from the connector menu works reliably again.
• The agent detail header no longer crowds on smaller screens.

One place to work

Command is now a single, consolidated surface: a quieter Home, a readable work rail, and Recent work as thin rows you can scan at a glance.

Composer sigils

Type @ to attach context, ! for run options, and # to save a memory — three characters that keep your hands on the keyboard.

Tasks, natively

Tasks are now first-class objects: assign one to a coworker and track it from chat or the task page.

Improvements

• Sign in with Microsoft or email, not just Google.
• Publish and update skills right from the Command Center.
• Install a playbook as a reusable personal skill.
• New connectors: Luma events, and personal Granola meeting notes your coworker can draw on.
• We rebuilt our database infrastructure — the app loads faster and live updates arrive sooner.

Fixes

• The composer menu no longer overflows on small screens.

Least-privilege connecting

When you connect an app, Supercenter now asks for the narrowest permissions by default — you can widen the scope if you want, but you start small.

Know which account you used

Connection cards show the identity you signed in with and offer to reconnect with the same one, so you're never guessing which account an app is on.

Improvements

• Connect options for apps with multiple sign-in methods now live in one tidy card menu.
• Your Command history is yours: press ↑ to recall a prompt, or browse it with ⌘K.
• New use-case pages for private equity and VC fund operations.
• The site now answers AI assistants with clean markdown — share a link and their agent reads it properly.

Fixes

• Custom MCP servers tell you when they need OAuth and let you reconnect in place.
• Removed custom MCP servers no longer linger in the connector list.
• The navigation bar stays visible on light-background pages.

Sessions that recover

If your sign-in expires mid-session, the app recovers in place instead of dropping you on a dead-end error page.

A clearer model picker

Choosing a model now shows provider branding and pricing more clearly, with better context controls.

Improvements

• New agents start on a smarter default model.
• A new agent plugins layout makes it easier to browse and install skills.
• Better search and exports in the Affinity connector.

Fixes

• Malformed tool calls no longer stall a run — they're recovered automatically.
• Meeting-note date filtering returns complete results.
• Emoji and non-Latin text no longer get clipped in agent messages.

More reliable custom MCP servers

Custom MCP servers handle OAuth properly now — they tell you when a sign-in is needed and let you reconnect without leaving the page.

New connector: Microsoft Advertising

Read your ads data and take guarded write actions from any agent.

Improvements

• Connected-account discovery is faster and more dependable for teams with many accounts.
• New case study: how TimeTac runs finance operations on Supercenter.
• Browser-only analytics setup for teams that don't want a tag manager.
• The integrations and trust pages read better on phones.

Assess your company in an afternoon

The new AI-readiness flow lets any team run a structured assessment — including a short voice interview — and get a scored, evidence-based report. No sales call required.

Improvements

• Agent pages now show when an incoming event was dropped, so silent failures become visible.
• Coworker skill assignments stay current when a new skill version publishes.
• Mobile polish across the site and the product preview.

Fixes

• The installed app no longer draws under the phone's notch or zooms inputs on focus.

Evidence-based assessments

AI-readiness reports are now grounded in what your team actually said — every score links back to evidence from the interviews.

Agents with their own keys

Agents can hold their own app connections, scoped to exactly what they need — no more borrowing one person's credentials.

Improvements

• Voice interviews work in your language and send verified email invitations.
• A more compact setup layout on the dashboard.
• New connector: Synthflow, for AI voice calls, transcripts, and campaign analytics.

Fixes

• Scheduled runs that kept failing now pause themselves and tell you why, instead of retrying forever.
• Agent owners get one clear notification when a run needs attention — no alert storms.

Share links

Share any Command chat with people outside your workspace using a public snapshot link — perfect for showing a colleague exactly what an agent did.

Supercenter on your home screen

Install Supercenter from your iPhone's Home Screen; it opens the full app, not a cut-down shell.

Custom MCP, your way

Custom MCP servers now support bearer tokens and OAuth, not just API keys.

Improvements

• Agents can read PDFs visually — charts and layouts included.
• Command home suggests starter prompts based on what you're working on.
• New platform pages on the site walk through what the product actually does.

Fixes

• Connect buttons show the error instead of silently doing nothing.
• Shared chats carry their starting context correctly.

Vision

Agents can now analyze images — screenshots, scans, and photos become inputs they can reason about.

Stop, politely

Hover a running session to reveal a stop button and cancel it cleanly.

See where the work goes

A new usage dashboard shows agent activity across your org with filters and the numbers that matter.

Improvements

• When a chat looks like a recurring job, Command offers to set it up as a scheduled run.
• Skill activity now appears in the admin usage feed.
• Uploading a skill archive gives clearer feedback — unsupported files say why.

Fixes

• Admins and owners can read run details again.

Our own sign-in

Sign-in and sign-up are now fully ours — including choosing an organization, creating a new one, and multi-factor authentication.

Free while in beta

Pricing is gone for now — everyone onboards free while we're in beta.

The blog

We launched a blog: field notes on AI coworkers, agents, and connected apps.

Improvements

• The Google Chat app is ready for the Workspace Marketplace.
• The homepage hero now shows a live Slack thread instead of animations.

Fixes

• Oversized tool results no longer overwhelm a run — large results are written to files and summarized.

Command, redesigned

The Command Center is now built around chat: one place to talk to agents, watch runs, and manage work.

A coworker per channel

Assign different coworkers to different Slack channels — each channel greets people by naming the coworker who covers it.

Custom connectors, no code required from us

Your org can now build its own connectors — API wrappers defined in the product and usable by every agent.

Improvements

• Run timelines show real summaries, connector icons, error messages, and permalinks to the Slack replies a run posted.
• The sidebar's Recent list is now one quiet line per chat.
• Personal coworkers show up in the coworkers area.

Fixes

• Fixed the app shell crashing for members without agent access.
• Fixed connect links dead-ending — you can now re-authorize from the link itself.

Custom MCP connectors

Add any MCP server by URL and it appears in your connector grid like a first-party app — for you or your whole org.

A calmer coworker

Your coworker is more careful about which messages deserve a reply, and ambient pickups show a thinking cue so you know it's working.

Improvements

• Slack-first onboarding: install the app, pick a channel, meet your coworker.
• An activity feed shows what your coworker has been doing in your channels.

Fixes

• Fixed duplicate replies in busy Slack threads.
• Scheduled agents with invalid settings pause instead of looping.

Coworker email

Your coworker now has its own email inbox. It can read incoming mail and draft replies — every send waits for your approval.

Run agents on your own model account

Teams can run their coworker on their own model key for full control over spend and limits.

Improvements

• Clearer states for delegated access — you can always see who granted what.
• Marketing pages now say "Start for free" and pricing lives on its own page.

Fixes

• Switching between organizations no longer signs you out.

A coworker with a name

Your Slack app is now a coworker: it answers as a consistent persona, shows a thinking indicator while it works, and replies in threads like a teammate.

Memory that sticks

Your coworker learns rules worth remembering — must-not-forget instructions persist across every conversation, and it can learn ambiently from conversations it only observes.

Personal coworkers

Everyone can have their own coworker with its own page, its own tasks, and a way to invite teammates.

Personalities

Give your coworker a persona — pick a curated preset or describe in plain language how it should sound and what it should never do.

Improvements

• Microsoft Teams joins Slack and Google Chat as a channel.
• The Trust Center is live — our security controls, subprocessors, and compliance posture in public.
• Task pages show everything your coworker has been asked to do.

Fixes

• Slack replies stream in smoothly instead of appearing all at once.
• Switching organizations no longer bounces you back to sign-in.

Approve in Slack

When a coworker needs a decision, it asks right in the thread with buttons — approve or decline without opening the app.

New connector: DocuSign

Agents can work with DocuSign envelopes through a dedicated connector.

Tasks and triggers

Give your coworker standing tasks and trigger them on schedules or events.

Improvements

• Built-in web search is now a default tool for every agent.
• Upload files to Slack through your coworker, and download attachments from it.
• Onboarding now starts with a chat that sets things up for you.
• Promotion codes work at checkout.

Three-step onboarding

New workspaces get a guided three-step setup — connect an app, pick an agent, run it.

Your accounts, plural

Connect multiple accounts for the same app and pick which one each agent uses.

Improvements

• Ready-made Slack agent apps you can install in one click.
• Prepaid credits are now a billing option.
• Workspaces can be deleted and recovered.
• Sidebar navigation preloads, so moving around the app feels instant.

Fixes

• Slack follow-ups stay in their thread instead of wandering to the channel.
• Reconnecting an app cleans up the old connection properly.

Helpers in threads

Agents can spawn helpers that work inside Slack threads — answers land where the conversation is, and matched replies get an automatic reaction.

Replay a trigger

Re-run an agent on a past trigger event right from its page — useful for testing, and for recovering from a bad run.

Improvements

• Set a cost budget per run so long jobs can't surprise you.
• Invite whole teams at once with bulk invites.
• Pick which connectors each agent is allowed to use.
• Runs carry a step timeline so you can see what happened, in order.

Fixes

• Scheduled agents no longer occasionally fan out duplicate runs.

Permission policies

Admins can now set policies on what agents are allowed to do — restrict agent access by role and group, and keep routine runs off approval prompts.

An operations dashboard

A new dashboard shows agent health, spend, and an audit trail of admin overrides.

Improvements

• Skills can be shared with specific groups, and agents can use private skills.
• Agents always run the latest published skill version.
• Delegate specific actions to a teammate's connected accounts — explicitly, and revocably.

Fixes

• Pending-invitation lists paginate correctly, and invite emails say what access they grant.

Agents that run in the cloud

Agents now run in an isolated cloud environment — they can work on your connected apps for minutes at a time without holding your laptop open.

Wake on events

Agents can start automatically when something happens in a connected app — a new email, a form submission, a calendar change.

Improvements

• Connect MCP servers straight from the connector card.
• An admin usage dashboard shows what your org's agents are doing.
• Google Drive knowledge can feed your agents.
• The site got a fresh coat of paint — and a platform page.

Improvements

• Agents find the right action faster — smarter, fuzzier search across apps, skills, and people.
• Agents can download files from connected apps and read PDFs.
• Invite and manage teammates straight from your AI client.
• A new self-serve browser makes it easy to discover and connect apps.
• Your org can set shared context — a short brief every agent sees.

Fixes

• Member management handles pending invites and re-invites correctly.

A real app shell

Supercenter now has a proper home: a left sidebar, a minimal top bar, and everything where you expect it.

Everything is a run

Sessions and agent work are unified under one concept — a run — with one place to see what happened.

Agents with borrowed keys

Agents can act through a teammate's connected accounts — with that person's explicit approval, and a record of every use.

A Trust Console

Admins get a dedicated console for overseeing cloud agents.

Improvements

• Spend caps and billing for cloud agents, with an org-level margin option.
• Org-level credentials: connect an app once for the whole company.
• Upload skills as .zip or .skill archives.
• Our marketing site is live.

Fixes

• Connections no longer silently expire — tokens refresh ahead of time.

Files in, files out

Agents can upload files to connected apps and pull files back — reports, exports, attachments.

API keys

Programmatic access with API keys for teams that script their own clients.

Improvements

• The dashboard was rebuilt in the new design language.

Fixes

• Clearer errors when a connection fails — no more cryptic raw messages.

Improvements

• Supercenter now notices when a connected app's capabilities have changed and tells you to refresh — instead of quietly serving outdated actions.

New connectors

Crisp for support conversations, write actions for Affinity, and a Postgres connector for your own databases.

Connect any MCP server

Point Supercenter at any remote MCP server — OAuth and bearer auth supported — and its tools become available to your agents.

Improvements

• A monochrome design overhaul across the app.
• Onboarding now shows your MCP URL and paste-in instructions for your AI client.

Fixes

• Connections no longer ask you to sign in again nearly as often.

Hello, world

Supercenter is live: one secure gateway that lets AI assistants act on your company's connected apps.

The gateway

Connect apps like Gmail and Slack once, and every AI client you allow gets safe, audited access through a single MCP endpoint.

Skills

Package how your team works into skills — personal or shared — that agents can install and use.

The console

An admin console for managing skills, reviewing feedback, and onboarding your organization.

Improvements

• Admins and viewers get separate levels of access, and you can belong to multiple organizations.
• AI clients get a helpful orientation the moment they connect.