What's new.

The latest improvements and updates from Tandem.

Choose whether your agent can use a sandbox

You can now turn sandbox access on or off from Capabilities. It starts off until you enable it.

What changed

  • Switch off sandbox access to block commands and repository changes.
  • Channels can disable inherited access, but cannot override a workspace that has it switched off.

Start repository tasks with less waiting

Tandem spends less time downloading repository history before starting your coding task.

What changed

  • New tasks download the current branch snapshot to reduce setup time for repositories with long histories.
  • Checkout failures distinguish a timeout from authentication, certificate, or transfer problems.

Give longer coding tasks time to finish

Repository work can continue through longer setup and validation runs without being cut off after fifteen minutes. Your configured task timeout and approval rules still apply.

What changed

  • Increase the task step budget in Advanced repository settings for work that needs more research, edits, and checks.
  • Retry interrupted runs without losing unpublished work before the task deadline.
  • Simultaneous shell requests wait their turn instead of failing as busy.

Give larger repositories time to open

Coding tasks allow more setup time for repositories with large files.

What changed

  • Repository downloads can run for up to five minutes, within your configured task timeout.
  • Cancelling a download also stops its upstream transfer.
  • Coding tasks have more CPU, memory, and disk space for installing dependencies and running repository checks.

See where Tandem can work on GitHub

Your repository settings now show connected GitHub accounts and repository names as soon as you open the page, along with their Slack channels and pull request target branches.

What changed

  • Browse available repositories under each connected account and see when access needs attention.
  • Search channels when choosing where people can request repository work.
  • Find the coding task timeout under Advanced settings, with a clear explanation of what it limits.

Stop coding tasks reliably and recover interrupted work

Cancelling a coding task now stops work that was still waiting on a command or publication response. Already published changes remain available.

What changed

  • Cancellation and time limits are checked before each further publication step.
  • A late publication response no longer changes a cancelled task back to completed.
  • Restarting after a sandbox expires clearly reports when unpublished work was interrupted.
  • When someone changes the task branch, Tandem reports the conflict instead of asking you to fix permissions.

Start coding tasks reliably

Fixed sandbox startup and secure repository checkout failures when beginning a coding request.

What changed

  • Coding commands use the execution identity required by the sandbox.
  • Private repository checkouts trust the secure connection used by Tandem.

Recognize Tandem throughout your workspace

Tandem now uses the same name across settings, Slack replies, approval requests, and connection sign-in screens.

What changed

  • Approval requests and failure notices clearly identify Tandem.
  • Connection screens and help text use the same name as your workspace assistant.

Connect a GitHub app you already installed

If Tandem is installed on GitHub but your account is missing from repository settings, you can now finish connecting it without reinstalling the app or changing its repository permissions.

What changed

  • Choose “Connect installed GitHub app” in repository settings to finish linking an existing installation to your workspace.

Find the right settings for coding requests

When coding access is unavailable, Tandem now points you to its repository settings and explains the permissions needed to get started.

What changed

  • Coding setup guidance no longer asks you to connect an unnecessary GitHub MCP server.
  • Tandem explains that channel access and workspace approval rules still apply.

Open connected repositories for coding

Fixed a connection failure that prevented coding tasks from opening an enabled GitHub repository.

What changed

  • Connected repositories can now be checked and cloned when starting a coding task.

Request code changes from Slack

You can enable a GitHub repository in selected Slack channels and ask Tandem to make a change. Tandem reads the project’s instructions, edits the code, runs checks, and can publish a draft pull request with the results.

What changed

  • Follow up in the same thread to continue the work. Repository settings let you choose channels, a starting branch and directory, a run limit, and whether publishing requires approval. Existing workspace approval rules still apply.

Clearer coding connection failures

When a connected repository cannot be checked, Tandem reports the failure instead of incorrectly saying no repositories are enabled.

What changed

  • Coding connection failures no longer appear as an empty repository list.

Use your connected model without extra policy setup

Selected models can now run without a separate data-retention declaration in Tandem.

What changed

  • Removed retention settings that could silently block connected models, including AWS Bedrock.
  • Simplified provider setup so you only supply credentials and the provider’s required connection details.
  • Existing model selections work without reconnecting the provider or changing governance settings.

Approve connected tools from Slack

Connected tools that need your approval can now send their approval request instead of failing before the request reaches you.

What changed

  • Fixed an error that prevented MCP and browser approval requests from starting.
  • Repeated approval requests continue using the same pending request.

Reconnect a model provider after removing it

You can remove and reconnect AWS Bedrock or another model provider without an unexpected conflict error.

What changed

  • Reconnecting a removed provider now works even when its previous model catalog is still in your workspace history.

Read clearer answers in Slack

Tandem's Slack replies now display formatted headings, bold text, tables, lists, and code instead of raw formatting characters.

What changed

  • Short questions get more concise answers, with longer reports reserved for requests that need them.
  • Replies after approval stay focused on your original question.
  • Link previews no longer interrupt the layout of an answer.

Read skill files clearly

The skill editor now uses a consistent dark background with readable text and syntax colors.

What changed

  • Fixed pale text appearing on a white background when editing or viewing skill files.
  • Kept the editor usable when browser zoom leaves less vertical space.

Finish connecting MCP tools after sign-in

MCP connections can now finish loading their tools after a successful sign-in.

What changed

  • Fixed a connection error that blocked discovery after you authorized a server.
  • Retry tool discovery using your saved sign-in instead of repeating the login flow.

Sign in and finish connecting your MCP servers

Connecting an MCP server now opens sign-in when needed and loads its tools automatically after you return to Tandem.

What changed

  • Finish setup for servers that were previously left at “Added.”
  • See when a server is connected, retry a failed connection, or reconnect an expired sign-in.
  • Use an API token when a server requires one, and go directly to agent access settings once its tools are ready.
  • Connect older remote servers as well as servers using the current connection protocol.

Keep supported MCP connections signed in

Connecting an MCP server can now request permission to renew its sign-in automatically when the server supports it.

What changed

  • Servers such as Ally can provide renewable access instead of requiring another login when the first session expires.
  • Existing connections need one reconnect to grant this permission.

Connect tools when optional context is unavailable

MCP tools can finish connecting when a server advertises optional context sources that it does not provide.

What changed

  • An unsupported context list no longer prevents available tools from connecting.

Start MCP sign-in without a connection error

Connecting or retrying an MCP server now reaches its sign-in page instead of immediately showing a connection error. This fixes automatic sign-in across remote MCP providers, including PostHog.

What changed

  • Retry existing connections to open the provider’s sign-in page.
  • Connected tools run without incorrectly rejecting valid arguments or structured results.

Approve requests to connected tools

Requests to connected MCP tools no longer fail with an approval mismatch before you can approve them. This applies to PostHog, Ally, and other connected servers that require approval.

What changed

  • Approval requests can now proceed to connected tools after you approve the exact action.
  • Unapproved or changed actions remain blocked.

Approved tools can finish in Slack channels

Approving a tool request from a Slack channel now lets Tandem carry out the approved action and return its result.

What changed

  • Channel approvals no longer fail with an incorrect configuration-change message when the configuration has stayed the same.
  • Approval still applies only to the exact action you reviewed.

Connect Bedrock without a false validation error

Connecting or refreshing Amazon Bedrock no longer fails when the model list ends with an empty continuation token. Unexpected provider responses are now reported as provider errors, instead of suggesting that your connection form is malformed.

What changed

  • Connect and refresh Bedrock even when the provider explicitly marks the end of its model list.
  • See a provider error when Bedrock returns an unexpected response.

Find more Claude models through Amazon Bedrock

Your Bedrock model list now includes Claude Sonnet and Opus options available through cross-region inference. Refresh your connected provider's models to see the expanded selection.

What changed

  • Browse models across all pages of Bedrock's available inference profiles.
  • Use supported Claude and Nova models with agent tools.
  • Get a clear permission error when Bedrock prevents the complete model list from loading.

Let routine actions proceed automatically

You can now set an approval policy in General so Tandem can handle permitted reads and clearly requested, small reversible changes without interrupting you.

What changed

  • Choose Auto, Ask before changes, or Read only, and write additional approval instructions.
  • Keep consequential or uncertain actions in the existing approval flow in your Slack thread.
  • Read only blocks changes and uncertain calls without offering an approval that cannot run.
  • Verify read-only MCP tools to let them run without a separate model review. A changed tool catalog requires verification again.
  • Existing approval restrictions stay in place until you change them. Automatic changes have per-request and workspace limits.

Approve actions in the conversation where you asked

Approval cards now appear in the same Slack conversation as your request, keeping the decision and result together.

What changed

  • Channel requests receive an approval card in their reply thread, including requests that start as a new channel message.
  • Approve or reject a single action without switching to a direct message. Only the requester can decide, and requests still expire after 30 minutes.
  • Direct-message requests keep their approval cards in that direct message.

Use your published skills in conversations

Enabled skills now load their published instructions without an incorrect checksum error.

What changed

  • Fixed activation of valid skill files while preserving checks that reject altered files.
  • Fixed approval requests being marked as failed after Slack had successfully delivered them.

Connect your Slack workspace from one button

Continue with Slack now guides any Slack workspace through installation and sign-in, while connected workspaces go straight to their workplace.

What changed

  • Each workplace connects its own Slack workspace and keeps its access separate.
  • Installation continues automatically into sign-in with the same Slack account.
  • Cancelled or expired sign-ins explain how to try again, including when workspace admin approval is needed.

Give your agent access to public websites

You can enable web fetching and browsing separately, so your agent can read public pages, extract information, and interact with websites.

What changed

  • Choose web access in the workspace agent’s capabilities and disable it for individual channels.
  • Read public websites without connecting a separate MCP server or signing into a website.
  • Let your agent click, fill forms, and submit actions using your workspace’s approval policy.
  • Existing workspaces keep web access off until an administrator enables it.

Connect private GitHub Skills to Tandem

You can now give Tandem read-only access to selected private repositories and import a Skill by pasting its GitHub URL.

What changed

  • Tandem checks public and already-authorized private repositories from one URL field.
  • If access is missing, Tandem guides you through granting access to only the repositories you choose.
  • Linked Skills can stay current with repository changes while preserving review and rollback controls.

Import any GitHub Skill from one URL

Paste a GitHub repository or Skill URL and Tandem now checks the access you already have before asking you to configure anything.

What changed

  • Import public repositories and already-shared private repositories through the same URL field.
  • Get a direct **Give Tandem access** action only when a private repository has not been shared yet.
  • Return from GitHub authorization with the pending repository URL restored.
  • Keep repository-access management separate from the import flow.

Connect private GitHub repositories as Skill sources

Workspace managers can connect GitHub, browse repositories and branches granted to Tandem, and copy or link Skills from private repositories. Linked Skills can require review or automatically apply valid upstream revisions, with pause, history, and rollback controls.

What changed

  • Import public GitHub Skills without connecting an account.
  • Browse only the private repositories and branches granted to Tandem.
  • Choose reviewed or automatic updates and safely return to an earlier version.

Keep navigation in place while creating a Time Loop

The Time Loop form now scrolls inside the app without moving the whole page or leaving empty space below it.

What changed

  • Workspace navigation and your user menu remain available when you enlarge text or zoom in.
  • Moving through form controls keeps scrolling inside the working area.
  • New pages open at the top, while going back restores your previous position.

Run trusted work automatically with Time Loops

Time Loops can now wake up on their schedule, use the selected model and workspace capabilities, and show what happened from admission through Slack delivery.

What changed

  • See whether a Time Loop is ready before activation, with clear actions for missing models, connections, channel access, or Slack permissions.
  • Choose whether each scheduled Slack message needs approval or explicitly authorize recurring posts to one selected channel.
  • Follow real run history, progress, approval state, failures, and confirmed or unconfirmed delivery outcomes.
  • Pause, update, or archive a Time Loop while preserving its schedule, immutable run configuration, and history.
  • Keep scheduled Slack approvals and delivery working when Slack rotates the app's access token.

See when your mention is being handled

Tandem adds an 👀 reaction to an accepted direct mention before preparing its reply, so you can see that your request is being handled.

What changed

  • Direct mentions receive a quick acknowledgement, including mentions inside a thread.
  • Disabled channels stay silent, and a failed reaction does not prevent a reply.
  • Existing installations can grant reaction access by reconnecting Slack.

Choose channel access for each Time Loop

You can now select which Slack channels a Time Loop may read and which channels it may post to when creating or updating its configuration.

What changed

  • Set reading and posting permissions separately, with no channels selected by default.
  • Search channels and review access before saving a loop’s changes.
  • Unavailable channels cannot receive new permissions.

Recognize people in channel conversations

Channel settings and Sessions now share a consistent page layout, with Channel settings opening first. Conversations show available Slack names and avatars, readable mentions, and more space below the agent workflow.

What changed

  • Keep the same channel header and tabs when switching between settings and sessions.
  • Recognize people by their available Slack names and avatars, including readable mentions.
  • Scan agent workflows more comfortably with extra space before progress and details.

Get Slack replies from your configured model

Your agent can now use enabled models when responding to Slack mentions. A setup problem no longer prevents the agent from sending its diagnostic reply.

What changed

  • Models selected in Agent settings are available when a Slack mention runs.
  • Slack replies no longer fail before reaching Slack.

See when an agent request fails

Requests that fail in the background now show a failed status instead of remaining submitted. When Slack is available, Tandem also posts a clear failure notice in the conversation.

What changed

  • Background failures no longer leave requests showing as submitted.
  • Failed requests show a clear notice when Slack can receive it.

Choose models from your connected providers

You can now browse and search the models discovered from your connected providers, then enable the models you want your agent to use.

What changed

  • Connected providers show their current model catalog, including context limits and tool support.
  • Enable a model to make it available in Agent → Models.
  • The agent’s empty model screen explains how to enable a model from an existing provider.

Read your agent’s model order clearly

Your agent’s primary and fallback models are easier to read in Agent → Models.

What changed

  • Primary and fallback labels no longer overlap model names.
  • Long names wrap within the available space on smaller screens.

Find your enabled models quickly

You can now show only enabled models and search within that list. Model rows scroll inside the catalog while search, filters, and page controls stay in place.

What changed

  • Switch between all models and enabled models with one click.
  • Keep your place on the Models page while browsing the catalog.
  • Open Providers to manage connections and refresh their models.

Follow agent work within a channel

You can browse a channel's sessions and see retained conversations alongside expandable agent updates.

What changed

  • Check execution, pending approvals, and Slack delivery separately in one feed.
  • Open recorded progress when you need more detail about a long-running request.
  • Channel settings opens first. Session history loads only when you open Sessions, with compact rows and scrolling inside the view.

Your configured agent can respond immediately

An enabled agent with a selected model can now respond in Slack without a separate activation step.

What changed

  • Hidden execution switches no longer prevent configured agents from responding.
  • Workspace and channel settings continue to control where the agent runs.
  • Capability permissions and tool approval requirements remain in effect.

Get help safely in the Slack conversations where you work

Tandem can now respond to direct requests in Slack using the exact skills, connected tools, data rules, and model choices configured for that workspace and channel.

What changed

  • Ask Tandem in an enabled channel or in a one-to-one direct message and receive the response in the same conversation.
  • Channel-specific skills and connected tools stay isolated from other public, private, and direct-message conversations.
  • Actions that need confirmation wait for the original requester to approve or reject the exact action in Slack.
  • Duplicate Slack deliveries and uncertain external outcomes are handled without blindly repeating an action.

Connect model providers with fewer setup decisions

Model provider setup now chooses the appropriate connection path automatically, so you only need to enter the credentials and provider-specific details required to get started.

What changed

  • Removed the infrastructure routing choice from provider setup.
  • Automatically applies the supported connection path for each provider.

Review audit events without losing table context

The audit log now keeps its column headings and filters fixed while event rows scroll independently.

What changed

  • Moved the vertical scrollbar into the event rows area.
  • Kept columns and filters visible while reviewing longer pages of activity.

Spot model-provider actions immediately

Available model providers now look clickable before you hover over them.

What changed

  • Provider cards keep their bordered, lime-accented action treatment at rest.
  • Hover gives available providers a subtle lift.
  • Providers that have already been added remain visually quiet and clearly unavailable.

Get to skill options faster

The Add skill dialog now opens directly onto its available source options without a redundant introductory line.

What changed

  • Browse skills.sh, Import GitHub, and New skill begin immediately beneath the dialog title.
  • Input-specific guidance remains beside the fields where it is useful.

Switch skill sources without the dialog jumping

The Add a skill dialog now keeps a steady footprint as you move between browsing, importing, and authoring.

What changed

  • Removed the empty line beneath the skills.sh search.
  • Kept tab content in a consistent, scrollable area.

Choose a model provider with less clutter

Available model providers now appear directly on the Models page without an extra container or repetitive guidance.

What changed

  • Removed the redundant provider-picker heading and security note.
  • Kept each provider choice clearly separated and ready to add.

Read agent settings more easily

Agent setting descriptions now use the available space instead of wrapping into an unnecessarily narrow column.

What changed

  • Status labels and switches take only the room they need.
  • Explanatory text stays wider and easier to scan.
  • Dropdown settings retain their established control width.

Connect OpenRouter through Cloudflare AI Gateway reliably

Workspace owners can now validate an OpenRouter API key and import its model catalog while routing model use through Cloudflare AI Gateway.

What changed

  • OpenRouter credentials are checked with OpenRouter before the connection is saved.
  • Cloudflare AI Gateway connections can reach the configured gateway from the hosted application.

Recover from Slack sign-in setup problems

Slack sign-in now returns you to a helpful recovery screen when your workspace still needs installation or reauthorization, instead of leaving you on a technical error response.

What changed

  • Missing Slack installations now lead directly to the install-required guidance.
  • Expired credentials or missing permissions now lead to reauthorization guidance.
  • Workspace access problems now return to the sign-in screen with a clear access message.

Focus on sandbox availability

The agent capabilities page now explains when sandbox tools are unavailable without a distracting empty-state illustration.

What changed

  • The unavailable sandbox-tools state is now quieter and more compact.

Remove model providers reliably

You can now remove a model provider connection without an incorrect request-validation error getting in the way.

What changed

  • Deleting a model provider now completes normally and records the change in the workspace audit log.

Create Time Loops with a cleaner, more consistent form

The Time Loop creation page now keeps its sections aligned, gives permission choices more breathing room, and uses the same action styling as the rest of Tandem.

What changed

  • Removed the redundant schedule summary and future-feature message.
  • Simplified scheduling to one sentence and only shows weekday choices for weekly loops.
  • Aligned section headings and schedule controls to a consistent left edge.
  • Improved spacing between Skills and MCP permissions.
  • Explains what is still required when Create Time Loop is unavailable.

Set up and inspect Time Loops in a focused workspace

You can now create a Time Loop on its own full page, review upcoming occurrences before activation, and choose the Skills, MCP connections, and model it can use.

What changed

  • Build one-time, hourly, daily, or weekly schedules with clear local-time controls.
  • Review the next three occurrences and safety policy before creating the loop.
  • Open a dedicated run-history workspace with search, time filters, status filters, and deep links ready for recorded executions.

Work in a calmer dark interface

Tandem now uses a charcoal product palette with layered dark surfaces, clear gray text, and cool blue interaction cues.

What changed

  • Use the application comfortably in a low-glare dark interface.
  • Find selected navigation, controls, and primary actions through clearer surface contrast.
  • See Tandem clearly on the dark sign-in screen, and keep filled actions readable while hovering.
  • Scroll long pages without a contrasting track interrupting the application surface.
  • Work through Agent settings with compact header actions that match the surrounding controls and appear only when changes need saving.

Work in a calmer, cooler application

The application now uses a subtle blue-gray canvas and navigation frame, replacing the previous brown cast while keeping working areas crisp and easy to scan.

What changed

  • Navigate and configure your workspace against a cooler, more neutral backdrop.
  • Distinguish navigation, working surfaces, and controls more clearly without adding visual noise.

Recognize clickable controls at a glance

Buttons and icon actions now use consistent pointer feedback throughout the product, including inside dialogs.

What changed

  • Standard, trash, edit, and other icon actions show the pointer cursor when available.
  • Dialog buttons and close controls use the same feedback.
  • Disabled controls show that they cannot currently be activated.

Connect the model provider your workplace already trusts

Workspace owners can now connect OpenAI, Anthropic, AWS Bedrock, or OpenRouter with their own API credentials. Tandem validates access, encrypts the credential, and imports the provider's available models before the connection becomes active.

What changed

  • Replace credentials and refresh a provider's model catalog without recreating the connection.
  • Route OpenAI, Anthropic, and OpenRouter through an authenticated Cloudflare AI Gateway when the deployment enables it.
  • Keep AWS Bedrock regional and direct when using a Bedrock API key, with clear guidance about Gateway's separate AWS credential requirements.
  • See provider and Gateway billing responsibility before choosing a connection path.

Find workspace administration where you expect it

Slack installation details and reinstallation now live in General settings, leaving Home focused and easier to scan.

What changed

  • Removed the resource counters from Home.
  • Moved Slack installation status and reinstallation into General settings.
  • Simplified Agent governance guidance and refined channel and Time Loop controls.

Set Time Loop schedules with clearer controls

New Time Loops now use consistent dropdowns for cadence and time while running on one predictable UTC clock.

What changed

  • Replaced browser-native cadence and time menus with the same dropdown style used across the product.
  • Split run time into clear hour, minute, and AM or PM choices.
  • Removed the timezone field and consistently save new schedules in UTC.

Take the next step from empty Skills and MCP screens

Empty Skills and MCP areas now explain what is missing and place the relevant action directly beneath the message.

What changed

  • Distracting character illustrations no longer compete with the explanation.
  • Empty screens lead directly to the action that adds or manages the missing item.
  • Filtered results use the same predictable message-then-action order.

See a simpler Time Loop starting point

The empty Time Loops view now keeps the focus on what recurring assignments do and where to create one.

What changed

  • Removed the decorative empty-state illustration from the Time Loops library.

Edit skill packages in a roomier workspace

The skill package editor now fills the available workspace and keeps its file list and code scrolling inside the editor, so the controls you need stay in view.

What changed

  • Work across the full available width and height when editing a skill package.
  • Find the skill’s enable, disable, and delete actions together at the top of the page.
  • Keep package actions and publishing controls visible while navigating longer files.

Find documentation in the header

Documentation is now available beside the Open app action while the landing hero stays focused on installation.

What changed

  • Open Docs directly from the landing-page header.
  • On mobile, open Docs from a compact hamburger menu while Open app stays visible.
  • See one clear installation action in the hero.

Restore the Tandem changelog

The public changelog is available again from the site navigation.

What changed

  • Open Changelog from the desktop header or mobile menu.
  • Browse the public release history at `/changelog`.

Review workspace configuration and security activity

Workspace owners and administrators can now review a durable audit log of configuration changes, sign-ins, Slack installation activity, and access denials.

What changed

  • Search activity, filter directly within each column, and choose which columns stay visible.
  • Sort and page through results, then open an event to review its safe metadata.
  • Sensitive configuration, credentials, tokens, and content never appear in audit records.

Understand empty and failed pages at a glance

Tandem now makes it clearer when a workspace area is empty, filtered, or temporarily unavailable.

What changed

  • Empty workspace areas now include a friendly Tandem illustration and a relevant next step when one is available.
  • Filtered views use a distinct no-matches expression so they do not look like a brand-new workspace.
  • Pages that fail to load now show a clear error state with a retry action when retrying is safe.

Open Tandem with one smooth loading sequence

Tandem now moves from startup into your workspace without restarting its loading animation.

What changed

  • The opening animation now continues until the first page is ready instead of appearing twice.

Connect an MCP server in one clear step

You can now add and manage an MCP server as a single item without creating separate server and connection records yourself.

What changed

  • Connect a server with a recognizable name and its secure URL.
  • See connected servers, status, and discovered capabilities together in one list.
  • Disconnect a server with one action when it is no longer needed.

Find models, MCPs, and logs faster

The application now uses shorter, consistent names for Models, MCPs, and Logs across navigation, page headings, summaries, and status messages.

What changed

  • Find Models, MCPs, and Logs under the same shorter names throughout the application.

See how Tandem works with your team

The homepage now shows how one shared AI teammate can use company context, notice work that needs attention, and help inside Slack without crossing your company’s boundaries.

What changed

  • Follow a realistic Slack conversation from question to useful answer and approved follow-through.
  • See the context, proactive signals, permissions, and channel rules that guide Tandem’s help.
  • Understand setup, model-provider choice, self-hosting, and hosted pricing before installing Tandem.
  • Open marketing and documentation pages directly, without seeing the workspace loading sequence.
  • Read the homepage in a cleaner, more direct sans-serif type system.

Enjoy a seamless homepage header

The homepage header now blends cleanly into the main page background.

What changed

  • See one continuous background behind the navigation and opening message.

See more of the audit log

The audit log now uses the full available workspace width, and its column menu more clearly explains what you can show or hide.

What changed

  • The audit table reaches the edge of its workspace without leaving an unused strip.
  • The Columns menu now includes the same clear introductory label as the Time zone menu.

Set up recurring agent assignments with confidence

You can now create and manage Time Loops from a dedicated workspace view, preview the next three wake-ups before activation, and keep earlier definitions available as assignments evolve.

What changed

  • Create one-time, hourly, daily, or weekly agent assignments with an explicit timezone.
  • Review the upcoming schedule and safety guardrails before activating a Time Loop.
  • Edit details, publish a new assignment or schedule revision, pause and resume work, or archive a loop without losing its history.

Focus the landing page on Tandem

The landing page now presents one clear animated introduction with two direct next steps.

What changed

  • See only the Tandem hero, installation action, and documentation action on the landing page.
  • Use a simpler header without Product, Customers, or Changelog navigation.

Open and search large workspaces faster

Workspace pages now load only the information they need, so opening the app and moving between sections feels more immediate.

What changed

  • Navigation prepares likely destinations without repeating the same requests.
  • Channel and skill searches stay responsive as a workspace grows, with results loaded in manageable pages.
  • Skill editing opens the selected file first while preserving complete packages when you publish.

Open Tandem’s website and documentation faster

Public pages now arrive ready to read and navigate, even on slower connections or when browser scripts are unavailable.

What changed

  • See homepage, FAQ, changelog, privacy, and documentation content as soon as the page arrives.
  • Keep reading and following public links when JavaScript is blocked or fails to load.
  • Search documentation on demand without downloading the full search data during the initial page load.
  • Avoid downloading desktop hero motion on smaller screens or when reduced motion is preferred.
  • Open the Product menu by hovering or clicking, switch between menu sections reliably, and dismiss it by clicking elsewhere.
  • Keep only Open app and the menu button in view while scrolling, then open one simple dropdown with every destination whenever it is needed.
  • Preview the same responsive menus when running the website locally.
  • See Tandem and its connected tools animate directly inside the homepage introduction.

Get straight answers before Tandem joins Slack

Explore a new FAQ page covering how Tandem works, what it can access, how your company stays in control, and what it costs.

What changed

  • Jump directly to questions about context, privacy, models, tools, pricing, open source, and self-hosting.
  • Read every answer on one page without opening and closing accordions.
  • Move from an answer to installation or the documentation when you are ready.

Audit timestamps stay in their column

Long timestamps in the audit log now shorten cleanly instead of overlapping nearby event details.

What changed

  • Audit timestamps remain contained within the Time column at narrower widths and larger text sizes.

Keep navigation in place while working

Long workspace pages now scroll inside the main content area, keeping the application frame and navigation steady and within reach.

What changed

  • Page scrolling stays inside the white workspace surface instead of moving the entire browser page.
  • Workspace and Agent navigation remain fixed while reviewing long forms and configuration sections.
  • The same behavior applies on desktop and mobile layouts.

Choose settings from consistent dropdowns

Dropdown controls now look and work the same throughout the application, making configuration choices easier to scan and navigate with a mouse or keyboard.

What changed

  • Open every dropdown with the same clear selected state, checkmark, and focus treatment used by the audit log.
  • Keep the surrounding page fixed in place while opening or closing a dropdown.
  • Navigate configuration choices consistently across agent, model, MCP, channel, and workspace screens.
  • See disabled choices clearly without losing the surrounding configuration context.

Configure your workspace agent from one clear interface

Workspace owners and administrators can now manage the agent’s instructions, behavior, skills, connected tools, governance, models, and channel-specific choices without editing raw configuration.

What changed

  • Set reusable workspace instructions and control how the agent searches authorized Slack context.
  • Choose skills, connected tools, model routing, approval requirements, and data-handling rules from purpose-built controls.
  • Review each channel’s effective setup, then safely inherit or tighten its instructions, models, skills, tool access, and approval policy.

Make more room for workspace content

You can now collapse the workspace navigation into a compact rail whenever you want more room to work.

What changed

  • Collapse beside your workspace name, then expand from the first control in the compact rail.
  • Keep every destination within reach while the navigation is compact.
  • Keep the working column centered at its familiar width as the navigation changes size.
  • Continue using the full navigation menu on smaller screens.

Read and scan the app more comfortably

The workspace interface now uses a calmer, more consistent text hierarchy so pages are easier to scan without heavy labels competing for attention.

What changed

  • Page titles, navigation, tables, forms, badges, and dialogs now share a restrained type system.
  • Technical text remains visually distinct while ordinary labels use a clearer, more readable style.
  • Existing button feedback, including shadows, hover treatments, and press motion, remains intact.

Know when Tandem is opening your workspace

Tandem now uses a branded loading screen while it verifies workspace access on slower connections.

What changed

  • A fluid curve grows into the Tandem character—right through the hands and fingertips—then finishes with a big smile while your workspace opens.
  • Fast page loads continue without an unnecessary loading flash.
  • Once the character appears, Tandem completes the animation before revealing your workspace—even when loading finishes early.
  • Errors can now use a crying Tandem with tears falling at varied angles, while empty pages use a calm Tandem with only a subtle eye twitch.
  • The loading screen remains still when reduced motion is enabled.

Add model providers with one click

You can now choose a familiar model provider by name and logo without filling in internal connection details.

What changed

  • Add OpenAI, Anthropic, Google AI, or OpenRouter directly from the model-provider page.
  • Recognize connected providers at a glance from their official marks.
  • See the current Slack mark when signing in to Tandem.

Keep moving without losing the menu

You can explore Tandem’s website with more room for the story while the actions you need stay close by.

What changed

  • The main navigation now sits in a calm, centered position instead of hugging the edge of the screen.
  • As you scroll, the full header makes way for a compact app button and menu.
  • The same simple menu remains easy to reach on smaller screens.

Meet Tandem

You can now spot Tandem instantly wherever you work, with a friendly new character and a clearer, consistent identity.

What changed

  • The new Tandem logo now appears across the website, documentation, sign-in, and workspace experience.
  • On the main website and sign-in screens, Tandem now playfully follows your cursor while the documentation logo stays still and focused.
  • A matching browser icon makes Tandem easier to find among your open tabs.
  • The logo remains clear on both light and dark backgrounds.

Your AI teammate, all in one place

You can now set up your workplace AI and choose its boundaries from one clear, focused workspace.

What changed

  • Sign in with Slack and move directly into the workplace you are allowed to manage.
  • See your current Slack profile picture and workspace icon, kept in sync with Slack.
  • Choose the skills, connections, models, and channels your AI teammate can use.
  • Find step-by-step guidance for setup and everyday administration in the public docs.

Control MCP access for every channel

You can now choose which connected MCP tools and context sources are available across your workspace, then inherit or adjust those choices for each channel.

What changed

  • Turn an MCP connection on or off for the workspace or a specific channel.
  • Set each tool to run automatically, ask for approval, or stay unavailable.
  • Keep context sources private by default and include only the sources you choose.
  • See inherited channel choices clearly before adding an override.

See exactly what your AI can access

You can review how workspace defaults and channel choices come together before your AI teammate gets to work.

What changed

  • Review the resources and abilities available in each workplace context.
  • Check whether a connection is ready without exposing its private credentials.
  • Unavailable or unauthorized resources stay out of reach automatically.

A safer way to connect Slack

The first research preview gives each workplace a private, clearly bounded foundation for its AI teammate.

What changed

  • Connect Slack through a secure sign-in and installation flow.
  • Keep supported Slack activity separated by workplace.
  • Catch invalid information before it reaches your workspace experience.