What we shipped.
Newest first.

Give a conversation a goal. The agent keeps working until it meets the goal, so you do not have to type "continue".

Read any PDF

Agents can read PDF files, including scanned pages.

Ready-made workflow skills

New skills help agents with three common jobs: action items from meetings, action items from documents, and email inbox triage.

Rich attachments

Images, videos and PDFs that you attach show directly in the conversation. Audio files play inline, and you can open the email attachments that an agent read.

Sub-agents share files

Sub-agents can now receive files from their parent agent and send files back. This makes large batch work faster.

Clearer runs

When a run fails, the conversation tells you why. Click Continue to resume it. When a long conversation is summarized to save space, you can see this step in the conversation.

Smaller improvements

GPT-6.1 Terra and Terra Fast join the model list. Tools that can email outside people now ask for your approval on every call. The sidebar shows more project chats and recent chats with Show more. In the desktop app, right-click a chat for its actions and double-click to rename it.

Everything an agent needs from you is now in one place. Needs you has one tab for approvals and one tab for tasks. You can finish, dismiss or snooze each item to any time.

Mentions that reach people

When an agent mentions a teammate, the mention shows their name, and they get a notification. When you come back, d5s shows the mentions you missed.

Maps in chat

Agents can show places on an interactive map in the conversation, for example for a trip or a route.

Desktop tabs

The desktop app opens each workspace in its own tab. Drag tabs to change their order.

A better browser

Browser screenshots appear in the conversation. Agents can move files between websites and their workspace. Open the Browser panel to replay what an agent did earlier.

Workspace instructions

Owners and admins can set instructions that every agent in the workspace follows.

New connections and file tools

Connect any email server with IMAP and SMTP. Agents can also convert Office files and calculate Excel formulas. JSON files open as a tree, and you can zoom HTML previews.

Mark any part of a preview and tell the agent what to change. Request edits works on Word, PowerPoint, PDF, Excel, Markdown, HTML and image files.

Agents learn skills

Type /learn to save what you just did as a skill. When an agent learns or changes a skill, a notice shows it, and you can undo the change.

Skills are easier to organize

Upload skills as folders, zip files or single SKILL.md files. Related skills group into plugins, and you add a whole plugin to an agent with one click.

Project memory you control

You can now edit and delete project memory. Agents can search project memory by keyword.

Work with agents in Projects

Type @ and an agent name to ask that agent in a Project chat. Agents can also ask each other for help across shared Projects.

More connections

Agents can send email from Outlook mailboxes. Connect Datadog and PagerDuty with a token. The Google Drive connector now reaches shared drives.

Smaller improvements

Urgent items from an agent now reach your phone and watch as notifications. Agents can show web image galleries in their replies. The agent's plan and progress updates now show as normal conversation text, outside the list of steps.

The new d5s app for Apple Watch lets you give an agent a task by voice and see the answer on your wrist. The mobile app also supports dictation.

Wake agents with events

Under Autonomy, set event triggers for an agent. A merged GitHub pull request, a new release or any webhook can start the agent's work.

Interactive widgets in chat

Agents can show small interactive views in the conversation, for example a dashboard with filters, tabs and drill-down.

More video tools

Agents can edit existing video, copy the motion of a reference clip, and end a clip on an image you choose.

Meetings and Jira

Calendar auto-join now joins Microsoft Teams and Zoom calls as well as Google Meet. You can change the transcription language during a live call. Connect Jira Cloud to let agents find, create and update issues.

Desktop notifications

The desktop app tells you when an agent finishes work while the window is in the background. A badge shows the items that need your attention.

Skills and AWS setup

Add supporting files to a skill in the skill editor. Agents can also guide you through AWS setup in the conversation.

Open a live call to see the new meeting view. The agent writes the meeting notes during the call and keeps them up to date as people talk. When the call ends, the follow-ups appear in the summary card.

Video and image editing

Agents can now create short videos in the conversation. Upload a picture and describe a change, and the agent edits your picture instead of making a new one.

Charts in chat

Agents show line, bar, area, pie and scatter charts inside the conversation. Switch any chart to a table to see the numbers.

Browsers that stay signed in

Browser profiles keep their sign-ins between sessions, and you decide which agents can use them.

Connect your AWS account

Connect an AWS account to let agents check and manage your cloud resources from chat, Slack, Teams and email. Your own AWS permissions decide what each agent can do.

A cleaner workspace

The sidebar is now compact and shows the agents and projects you pin. Settings have a new layout, and dark mode is easier to read. Workspace admins can invite colleagues after they review a suggestion from the agent. On Mac, Command+N starts a new chat.

Every model picker now offers three levels: Light, Medium and High. Choose how much power a task needs. d5s keeps each level up to date with strong current models, such as GPT-6 and Claude Opus 5.5.

Edit Markdown files directly

Open a Markdown file from the conversation and click Edit. You change the formatted document in place, and it saves automatically. The agent uses your saved version in its next step.

One menu for @ and /

Type @ or /, or click +, to open the same menu in every composer. It finds files, agents, skills and commands. You can also drop or paste files straight into the composer.

A simpler Agent Builder

Describe the agent in the chat on the left. A live agent card on the right shows the result as you build it.

Connectors you can explore

Each connector now has its own page that shows what it can do. You can share the tool permissions of a connection with your team. HubSpot connects in one step and gives agents CRM tools.

Desktop file access and more

In the desktop app, give an agent access to a folder on your computer: read only, write only, or both. Agents can now read text in scanned images in English, Dutch and Greek. Telegram agents can send you a message first and exchange files in both directions.

You can now edit a message you already sent. Open Edit message, change the text and send it again. The conversation goes back to that point and continues from your new message.

Dictation in the composer

Click the microphone in the chat composer and speak. Your words appear as text. You check and edit them before you send.

Reach your agents in more places

Agents can now work in Discord and Microsoft Teams, with their own name and identity in each channel.

New connectors

Connect Microsoft 365 to let agents read SharePoint sites and OneDrive files. Azure DevOps gives agents read access to your repositories.

Meetings join themselves

Choose the calendar meetings that an agent joins automatically. More than one agent can join the same call. Call history shows each call and what happened after it.

Better documents and a live browser

When an agent edits a Word file, the preview shows its tracked changes and comments. Agents can also use a live browser in the conversation, and you can watch each step.

Email setup is easier to scan: incoming and outgoing controls are grouped by provider, mailbox addresses stay visible, and sender rules can be adjusted without disturbing the rest of the connection. New-email requests now make the selected mailbox and destination explicit.

Connected mailboxes can also create and update drafts and send approved messages through the same permission controls as other connector actions.

Images at a glance

Raster image artifacts now open as images in conversations instead of generic file cards, on both web and mobile. They keep their natural shape, remain accessible, and open the exact displayed bytes when you want a closer look.

Dashboards are now durable workspaces for reports, charts, tables, and connected datasets. You can switch between Chat, Split, and Dashboard views, refresh individual widgets, and invite an authorized agent to contribute without losing the board’s history.

Keep the team informed

Shared automations now support teams of people and named agents, with clear Keep informed and Act on findings choices. Run results keep their own history, while handover receipts show when a coworker has received a result and give people a direct path to review it.

New personal accounts now have a guided first visit built around interactive clay coworkers. The short journey introduces departments, artifacts, communication channels, and workspace setup before opening chat, while returning users and invited teammates go straight to their workspace.

Find the right connected tool

Connected tools are easier to discover too: the catalog names each tool’s surface and the connector directory can surface exact operations before you use them. This makes it simpler for an agent to choose a native workspace action when one is available.