Chats

Chats are the interactive work mode in d5s. A chat keeps one session while you and the agent shape the result through multiple turns.

If a chat client offers a retry for an unconfirmed message, retry the pending message before composing a replacement. Retrying a pending message does not create a duplicate and keeps its attachments.

A persistent agent can receive work from connected Slack, Telegram, and email channels in its continuing conversation. People who can open the agent can review activity from its currently connected Slack channels. Sending, replying, and cancelling Slack work still require the appropriate Slack identity and membership. A manual chat does not automatically become an external reply. Every agent-authored Slack message shows Sent by an AI agent.

Telegram teammate DMs are a beta for private messages and files. A teammate must confirm their Telegram identity in the d5s browser before their messages can reach the agent. The linked DM uses the named agent's existing persistent session, and supports files in both directions and agent-started messages to linked teammates. It does not add group conversations, general interactive buttons, or streaming replies. When an Ask-gated connector action pauses work, the agent can show a single-use approval control in that same linked DM. See Agents for setup and limits.

Delivery status distinguishes provider acceptance from confirmed receipt. Accepted by Telegram means Telegram accepted the request, not that a person received or read it. If the status is Delivery unknown, check the private chat before sending again to avoid a duplicate.

Start a chat from the workspace home without choosing a project. It can use workspace connectors available to ordinary chats. Choose an existing project when the work should use that project's instructions, files, memory, skills, and connector choices, including connections the project removed.

In the sidebar, select the page-and-pen New chat icon to the right of a pinned project's three-dot menu to start a chat in that project.

In the macOS desktop app, press Command+N while viewing a pinned project or one of its chats to open a new chat in that project. From an agent, a standalone chat, an unpinned project, or another page, the shortcut opens New chat in your current workspace. This also works while typing and with the sidebar collapsed.

A dedicated Teams identity brings mentions from activated standard channels into the same agent conversation. The agent's ordinary response stays in d5s. The agent posts to Teams only when asked. Installing the app alone does not activate every channel. Authorized viewers of the agent can read its currently bound Teams-channel activity while the Teams app stays installed, without linking their own Teams identity. Reading this history does not grant Teams sending, approval, or cancellation authority.

After d5s accepts your Teams request, an agent message in that thread shows Received while the work waits to start. It changes to Working when the agent starts.

If an accepted request is cancelled or cannot start, that same message shows the outcome.

When Stop is available, select it to interrupt that work. The message shows Stopping until cancellation is confirmed, then Stopped. Stopping does not undo actions the agent already completed, and any reply already posted stays visible.

If Teams does not show a final confirmation, open the agent's activity in d5s to check the outcome before trying again.

The first reply the agent posts uses that same message. Once the work finishes, it becomes an ordinary formatted Teams answer. Further replies may appear as separate messages. Questions and approval requests keep their answer and decision controls; a question or approval does not erase an answer already posted. If the work finishes without a Teams answer, the message shows the outcome and a link to the agent's activity in d5s.

If the original Received message is confirmed but a progress update cannot be confirmed, the answer may arrive as a separate ordinary Teams message. If sending the answer itself is uncertain, check the thread and the agent's activity in d5s before trying again.

The model menu offers three levels: Light for quick answers and simple tasks, Medium for everyday work with tools and files, and High for deep reasoning and long tasks. d5s picks the model and reasoning effort behind each level for your workspace. To pick a specific model, type in Search models at the top of the menu or scroll to the model list below the levels. Each model shows its input window, such as 1M. The choice applies to you in every workspace. While your organization's shared Codex subscription is on, the menu shows the levels only. On the Free plan the menu runs Light only: Medium and High show Available on Pro and Team, and there is no model list. Organization Owners and Admins see Upgrade to Pro at the bottom of the menu.

The model list groups the current available catalog models by publisher, with one version per model family. Picking a model saves its family: later turns use the newest version of that family available under your organization's access and settings. Older saved catalog choices follow the same rule. Current controls show the available version name with speed separately. Work already running keeps its selected version, and historical labels keep the exact model used. Choose reasoning effort within the model menu. Organization Owners and Admins can open Model settings from the menu to manage availability and the global Fast setting.

When Fast is enabled, a small lightning icon appears beside the chat model control. The tooltip shows whether the current model has a Fast version. The icon shows the organization preference, so it stays visible when a model has no Fast version. Auto and level labels remain visible as usual.

Where Auto is available, it picks Light, Medium, or High from your message and your recent exchanges with the agent in that chat, and uses a model available to your organization. Once work starts, the composer keeps Auto visible and adds the chosen level, such as Auto · Medium.

In a listed Project, the model control opens the same menu: levels first, then the searchable model list. On the Project page, a workspace Owner or Admin selects the model that new ordinary chats in the Project start with, for every member. Workspace default removes the override. Each chat keeps the model it started with. A later Project change does not change it. Inside a chat, anyone who can send messages there can pick another model for that chat only. Agents, automations, and dashboards continue to use their own model settings.

Attach files and long text

A message can hold up to 1,000,000 characters and up to 50 files. Paste or drop files into the message box, or choose Attach file in the + menu. When you paste more than 16,000 characters of text, d5s attaches it as a numbered text file (for example pasted-text-1_2026-10-29_15-04-05.txt) and does not put it in the message box. The agent reads the file. Remove the file to undo the paste. A shorter paste goes into the message box. If a message is too large for the model, the conversation shows a notice; attach the content as a file or start a new chat.

Choose who answers in a Project

Project messages use d5s by default. From the Project overview or an existing Project chat, type @ followed by part of an agent's name, then choose a match from Agents. The list shows up to six matching agents. Type more to narrow it. The selected agent mention stays inline in your draft. Your remaining text and files stay in place. Selecting another agent replaces the first agent mention; one agent answers each message. Remove the agent mention to ask d5s. A restored unsent draft keeps its selected mention.

The agent mention selects who answers that message. After sending, a new message without an agent mention uses d5s again. The answer appears in the same chat with the responder's name. A named agent uses its own model preference, shown as a read-only label. Workspace default means that preference is inherited.

If the connection drops during a send, select Retry original request. This retries the original message with its original recipient. Settle that request before sending a different card answer, draft, or widget request in the same chat.

Use a chat when

• the question is exploratory or changing;
• you expect to review intermediate results;
• a tool action may need approval;
• the work should remain attached to one conversation;
• you want to prototype a prompt before making it repeatable.

During a run

The conversation streams the agent's progress rather than waiting for one opaque response. Depending on the task, you may see tool activity, sub-agents, research phases, generated artifacts, connector approval cards, and a request for access to an exact external API host. When an allowed website sends a named agent to another website, the agent can show Allow a website? in chat. Review the exact hostname, reason, and Current browser access. A workspace admin can choose Allow and continue or Decline. See Agents for the scopes.

Open a file that the agent presented from its card in the conversation. Saved presentations, including HTML reports and CSV files, remain available after the coding environment stops. The card opens the saved version that the agent presented. Files in the sidebar show the current file while it is available. If those files have expired, Files opens or downloads the latest saved version that the agent presented in that conversation. Saved files remain listed even when their original messages are outside the loaded conversation history. Choose Load more saved files to browse further.

A CSV file opens as a table, with its row and column count above it. The table shows the first 1,000 rows and 100 columns. Select Show source to see the file as text.

On a shared workspace computer, each presented file can be up to 25 MiB. An HTML report can include up to 200 files and 25 MiB in total. Larger files are refused, not truncated. Creating a file alone does not present it: ask the agent to present the files you want to keep in the conversation. If saving is unconfirmed, keep the original files until saving succeeds.

For quantitative answers, ask the agent to chart the values in the conversation. Inline charts support trends, category comparisons, parts of a whole, and numeric relationships. Tap or click a chart to inspect a point in a compact detail panel below it; moving over the chart alone does not open the panel. Open the chart menu to View as table or Download as image. The table uses the same observations as the chart, including missing values. When the agent has source information, the chart can show its source, retrieval date, units, and notes about partial periods. An agent can chart a small result directly or prepare the data in its coding environment first; the displayed values are saved with the conversation.

For a quick snapshot, ask the agent to show two to six headline values as metric cards. Each value has a label and can include a short explanation. The cards use two columns in a narrow chat and show a source or retrieval time when the agent supplies one. The values remain visible with the conversation.

When a chat sends work to a named workspace agent, the chat continues without waiting for the agent's separate run. The request enters the agent's inbox. The requests, the answers, and the turns between them fold into one quiet line above the final answer, such as Worked with Research, Finance · 2m. While an answer is missing, the line reads Working with Research and the time so far. When work did not finish, the line adds 1 problem or stopped. Select the line, or press Enter or Space on it, to open each request with its answer, time, and state. Show full reply shows a long answer, Open Research's chat opens that agent, and Notes shows the text of the folded turns. Messages that other agents send to an agent show as one line under the agent's name in its next turn, such as from Research · Message, Reply. Select that line to open each message with its time, state, and Copy. The returned answer does not start new work.

To bring a colleague into the current workspace, ask the agent to help invite them and provide their email address. The agent can suggest a Member invitation, or an Admin invitation when you ask for that role. Its suggestion does not send an email or grant access. Review the named workspace, address, role, and any seat terms in the confirmation dialog, then select Send invitation. Invitation permissions and billing authorization still apply; see Sharing and permissions.

The approval control beside the + menu has two modes. Ask for approval pauses on connector actions configured to ask. Allow all lets those actions continue in this chat without individual approval cards. Gmail, Outlook, and IMAP / SMTP sends, new, changed, and cancelled Google Calendar and Outlook events, and Google Drive sharing changes and deletes still wait on a card that shows the exact message, event, or file (see Connectors). Explicit denials, account permissions, and organization or workspace restrictions still apply. MCP tools that send email or book meetings stay off under both modes. Switch back to Ask whenever you want the next such action to pause.

If an approval card is already waiting when you choose Allow all, d5s retries that action once. A failed retry leaves the card available so you can review or deny it manually.

An IMAP Save draft request uses the same email preview when its tool permission is Ask first. Choose Save to create an unsent mailbox draft. Saving never approves a send. Open your mail app to review and send it.

Cards for a Gmail, Outlook, or IMAP / SMTP send, a new, changed, or cancelled Google Calendar or Outlook event, or a Google Drive sharing change or delete show as one row each under the agent's reply for that turn. Select a row to open its card in place. When cards wait for you, the conversation header shows an inbox icon with the exact number of waiting items, up to 999. Select it to open Needs you. In an agent's conversation, it has two tabs. To approve lists every waiting card, oldest first. Tasks lists the agent's tasks that need you under Today, Tomorrow, and Later, by your local date. Questions, failed tasks, and overdue tasks are always in Today. Tasks with no due time are last in Later. In each group, questions and failed tasks come first, then the earliest due time. The panel opens on To approve when a card waits, else on Tasks. Select a row to open it in place. In d5s, only people who can edit the conversation can decide a card; others see its state.

The agent keeps one card for each request. When you ask for a change to a waiting email or event, the agent updates that card. The card moves to the agent's latest reply, so you see the change there. It then shows Updated by with the agent's name and the time, and an approval of the earlier version does not run. In an agent's conversation, you can also tell the agent that a waiting card is no longer needed. The agent then closes it, and nothing is sent. The card shows Closed with the agent's reason.

An external API request is not access by itself. The card shows the requested hostname and asks what access to grant for this chat. This website is selected by default and allows just the requested hostname. Choose All websites to cover public HTTPS websites for the entire chat; the button changes to Allow all websites and a warning shows that this is much broader than the request. Either grant permits ordinary HTTPS API methods and paths, can be revoked from the chat card, and ends with the chat. Clearing a persistent agent’s conversation also removes its chat approvals. Websites may retain data, perform external actions, or return prompt injection. Review the selected scope before approving, and never send credentials or other secrets.

Messages sent while work is active may join the current work or wait in a queue. Waiting messages show inside the composer, above the text box, in the order they will run. A message you send during active work shows Sending… there until it joins the queue. The queue shows three messages; choose N more to show the rest, and select a message to read it in full. Point to a message to show its actions. On a touch screen, the actions are always visible. Steer brings a waiting message to the front and asks the current run to stop, so that message can run next. Edit removes the message from the queue and puts its text back in the composer, so you can change it and send it again. The × deletes only the selected waiting message. You can also stop active work without deleting the conversation. If the agent reported that an approved video is processing in the background, that video can still finish; see approved videos.

To shorten older conversation context, type /, choose /compact, then submit the command. The timeline changes from Compacting context to Context compacted when it finishes. The command is available only for an existing, idle conversation; if there is not enough older context, d5s tells you. When a conversation becomes too long during a task, d5s compacts it automatically and the same row shows below the completed actions.

Work with Computer

If your conversation shows Computer, open it to view the desktop used for the task. A terminal calculation or file-writing task does not require you to open the viewer. Work in the same project can use the same desktop files and browser sign-ins; see Projects.

A file saved on the desktop is not automatically an attachment in the conversation. Ask the agent which files contain the result. A completed command and a presented file are separate results; check that the file card is available before relying on it as a saved attachment.

Use workspace tabs in d5s Desktop

d5s Desktop shows each open workspace in its own tab, in a strip at the top of the window. The strip stays visible when only one tab is open.

• To open a workspace in a new tab, select + or press Command+T (Ctrl+T on Windows and Linux). Choose a workspace from the list, or type to find one. If the workspace is already open, d5s selects its tab.
• To close the current tab, press Command+W (Ctrl+W). Closing the last tab closes the window.
• To go to a tab, press Command+1 to Command+8 (Ctrl+1 to Ctrl+8). Command+9 (Ctrl+9) goes to the last tab. Ctrl+Tab and Ctrl+Shift+Tab go to the next and previous tab.
• Drag a tab to change its position. Right-click a tab to duplicate, reload, or close it, or to close the other tabs.

When you open d5s Desktop again, it reopens your tabs. A tab that was in the background loads when you select it. If a page has unsaved changes, d5s asks before it closes that tab.

Get notified in d5s Desktop

d5s Desktop tells you when an agent run that you started ends while you work in another app. A notification shows when the run finishes, fails, or needs your decision, such as an answer to a question or an approval. Select the notification to open that conversation. Desktop does not show a notification while its window is in front, because the conversation list already shows the result. A blue dot on a background tab shows that a conversation in that workspace needs you.

The d5s icon in the Dock (macOS), taskbar (Windows), or a supported Linux launcher shows how many conversations need you. A conversation counts while it has a result you have not opened, or while it waits for your decision. The unread result clears after you view the conversation for a moment. A waiting decision keeps counting until you answer it.

Only runs that you started count, including your messages to a teammate's agent. A conversation with an unread mention of you also counts, and notifies you when its run ends. Other runs that teammates started do not notify you. While it runs, Desktop checks up to 10 of your workspaces, starting with the one you have open, so a run in another workspace still notifies you. A run in another workspace, or one that started elsewhere such as from Slack or a schedule, can take up to a minute to appear.

To change this, open Settings > Desktop > Notifications. Turn Agent notifications or the badge on or off for this device. Focus modes and Do Not Disturb on your computer also apply. On macOS and Windows, select Open system notification settings to manage them, or to allow d5s notifications if none appear.

Edit a Markdown artifact

Open a Markdown artifact (.md or .markdown under artifacts/) from its card in the chat or from Files in the sidebar, then click in the document. The editor opens with the cursor where you clicked. A click on a link opens the link, and selecting text does not start editing. You can also select More actions > Edit. When the current file is available, editing opens its latest contents, including changes made since the artifact was presented. You edit the rendered document directly; headings, lists, tables and links keep their formatting. d5s saves your changes to the file about one second after you stop typing. Select Done to return to the preview.

Before d5s sends your next message, it saves any pending edits. The agent then reads the file with your changes. If the save fails, the message is not sent and your changes stay in the editor; select Retry.

Editing pauses while the agent works in this chat and resumes when it finishes. If your last changes could not be saved before the agent started, the editor keeps them read-only so you can copy them. d5s does not save them over the agent's work. The same applies when the file changed after you opened it, for example because the agent wrote to it: d5s does not save your changes over the new version. Copy your changes, select Done and Leave, then click in the document to open the new version. Images show as their description and are not loaded. If a file has content the editor cannot keep, a note above the document says so and editing does not start; ask the agent to change that file.

Files uploaded to a chat stay with that chat. Project files and memory are reusable across work in the project. Use Save as project when the chat's context should become a named, listed workspace resource.

A persistent agent can read earlier messages from a connected Slack or Teams thread supplied with its current request. In an ordinary d5s chat, ask about the conversation itself or its searchable chat history. Existing channel permissions still apply.

For a persistent agent reached through Slack, attachments appear as references first. A request inside a Slack thread can also use accessible files shared earlier in that thread. The agent downloads only files it needs; downloaded copies remain available for later work and their contents remain untrusted. Ask the agent to attach generated files to its Slack reply. Attachment failures are reported even when the text reply succeeds.

Request edits on a file

Open a Word document, Markdown file, spreadsheet, image, PDF, PowerPoint deck, or web page (HTML) from its card in the chat. A deck opens as slides; wait until they show. Select Request edits in the file header, or press E. Move the pointer over the file: an outline shows the part that a click marks. Click a paragraph, list item, cell, line of text on a page or slide, or element of a web page to mark it. A click on an image marks the whole image. On a touch screen, tap to mark.

To mark a larger area, drag a box over it; on a touch screen, press and hold, then drag. A box snaps to whole paragraphs, list items, or cells. On an image, a PDF page, a slide, or a web page, the box stays where you draw it. While Request edits is on, links and buttons on a web page do not respond; scroll the page as usual.

Type a comment for the marked part and press Enter. Mark more parts as needed. Select a numbered marker to change its comment. Select Send to send all marked parts to the agent in one message. The agent changes only the marked parts and keeps the rest of the file. For an image, a PDF, or a deck, d5s also sends a picture of the marked area. For a PDF or a deck, it also sends the page or slide number and the text in the marked part. For a web page, it sends the marked page elements and their text, but no picture.

Select Cancel, press Esc, or close the file to stop. If you marked two or more parts, d5s asks before it discards them. Your message in the chat shows the file and each marked part with its comment.

Edit an image

Upload a PNG or JPEG, then describe the change you want. State what should stay unchanged, for example: "Change only the bottle cap to gold. Keep the label, lighting, and background." When image generation is available, the agent uses the supplied picture as the source and presents the edited result in the conversation.

For another revision, refer to the latest returned image and describe the next change. You can supply up to six source images in one request; identify which picture to edit and how any other references should be used.

Each source must be at most 3 MiB, 4096 pixels per side, and 4 megapixels. For a larger picture or another file format, ask the agent to prepare a separate supported copy first. An unreadable or invalid source must be corrected before the edit can run. Review the result before using it, especially text and details you wanted preserved.

Continue, branch, or repeat

Continue in the same chat when the new request depends on its history. Branch from a completed agent response when you want to explore a different direction without changing the original thread. Start a new chat when the work should not inherit the old transcript.

A chat run can take up to 100 steps. A step is one model response, which can include tool calls. When a run uses all of its steps before the task is done, the agent says what it finished, what is left, and where to pick up. A row under the reply reads Stopped after 100 steps, the limit for one run. Select Continue to send Continue as an ordinary message: the next run starts with the conversation so far. The button shows on the latest reply only, while no run is active, for people who can send in the chat. In Slack and other chat apps, reply continue. Branching is unavailable when the copied history includes named-agent context. Information already received from other Projects can stay in a branch; opening the original source still requires current access.

On the web, use the thumbs-up or thumbs-down icon on a completed agent response to mark it as helpful or unhelpful. Selecting an icon immediately saves your rating and keeps it selected even if you skip or close the optional details dialog. You can add reasons and a comment later, switch directions, or remove your own feedback.

In the mobile app, select the three-dot Message actions control below a response, or press and hold the response. Select Copy response to copy its text. Good response marks a completed response as helpful; select it again to clear that rating. Report response opens What was wrong? Choose Incorrect or unhelpful, optionally add details, then submit to save a negative rating.

To report concerning content in the mobile app, select Report response and choose Offensive or unsafe content. You can report a response while it is still being written. Review the selected response, add optional details, and submit. The response excerpt and your details are shared privately with the d5s team for review. Avoid including passwords or unrelated personal information. Long responses include only the first 20,000 characters; use the details field to identify the concerning part.

A report is separate from your helpful or unhelpful rating. Changing or clearing a rating does not withdraw a report. The receipt confirms that the report was received, not that review is complete. If submission fails, keep the sheet open and try again.

The conversation shows only your own selection. It does not show teammate ratings, aggregate counts, or other people's feedback details. Your rating does not change the agent's next reply.

Find people, context, and actions from the composer

Click + or type @ or / to open the shared composer menu. Search workspace colleagues, accessible projects, conversations and dashboards, platform and installed skills, and connected apps. Choosing a reference inserts its name and icon or avatar inline. A project, conversation, or dashboard reference does not copy its history into the current chat or grant access. In a Project, typing @ lists visible agents under Agents. Enter part of a name to filter them. Choosing an available agent inserts its mention and selects who answers that message.

Before you search, the menu shows core tools and connected apps. Typing a query also reveals matching colleagues, context, unconnected apps, and skills, with skills at the bottom. Selecting a skill or connected app references it in the draft; it does not enable or disable it. Use project settings for activation changes. If you can install connectors, searching an unconnected app shows Connect, which opens its setup flow instead of inserting a reference. Use Up and Down to move and scroll the highlighted row into view, Enter or Tab to select, and Escape to close. A surface only shows actions it can perform; unavailable commands explain why they cannot run.

Selecting Web search or Deep research enables that mode and inserts its icon and name inline in the draft. Removing the last inline selection for a mode disables it; undo restores both the selection and its mode. These tool selections are not separate badges below the message.

Palette rows keep the icon or avatar beside the name. Available agent rows show Ask in this project. Unavailable rows show Unavailable. Highlight or tap an unavailable row to read its full reason; it cannot be selected. Inline references remain part of the message text at the cursor; they are not badges, checkmarks, or on/off controls. In Projects, selecting an agent inserts an inline mention. Typing or pasting a name without selecting it does not choose an agent recipient. Remove the selected agent mention to ask d5s. Colleague mentions keep their existing notification behavior.

If you can mention colleagues, choosing one inserts the name as plain text; the mention becomes active when you send. You can only mention members who can already open the conversation: people without access to its project appear disabled in the picker, and mentioning never grants access. Share the project first if they need to see it.

A mention renders as the person's name in blue in the message and notifies them: the conversation pins to the top of their Recents list with an @ marker until they open it, and a notification toast appears if they are elsewhere in d5s. Opening the conversation clears the mention. Mentions of you are also highlighted, so the one highlighted name in a conversation is the one addressing you.

While you have unread mentions, a Mentions row at the top of the sidebar shows how many there are. Select it to see each conversation, and select a conversation to open it. If you were mentioned while d5s was closed, one toast tells you how many mentions you missed when you come back. Select Show all to open the list. This toast shows once per browser, not on every reload.

Agents can mention workspace members the same way in their replies, for example to flag a result for someone's review. An agent mention that targets someone without access to the conversation is shown as plain text and notifies no one.

When a conversation is connected to Slack, a Slack mention of a teammate who has linked their Slack account to d5s becomes a d5s mention. Unlinked Slack accounts keep showing as plain display names. To tag someone in Slack, the agent follows the channel mention rules in Slack setup.

Correct a sent message

Select the pencil beside one of your own messages, change the text, then select Save & send. Cancel leaves the conversation unchanged.

Messages covered by a saved context summary cannot be edited. Their pencil action is disabled with an explanation; send a new message to correct those instructions. Newer messages remain editable unless the edit would also replace an earlier summarized input.

Cancel any queued messages before saving an edit. If messages are still queued, the edit leaves them intact and keeps your correction in the editor.

Saving replaces that message and the conversation after it. The old text and dependent responses disappear from the chat and are excluded from future conversation context. If the agent is working, d5s stops it before submitting the replacement. Attached files stay attached.

Editing does not undo actions already completed, remove files the agent created, or reverse usage charges. You cannot edit messages received from external channels or replace a conversation tail containing someone else's messages or system instructions. Some older or branched messages cannot be edited if their original tool settings are unavailable. If saving fails, the editor keeps your correction so you can retry.

Find in a conversation

Press ⌘F (Command+F) on a Mac or Ctrl+F on other systems. You can also select Find in conversation (the search button) in the conversation header. Type at least two characters.

d5s searches your messages and the agent's responses in the whole conversation, including older messages that are not loaded yet. The bar shows your position, for example 3 of 17, and highlights the matches. The first match is the newest.

• Press Enter, or select the up arrow button, to go to an older match.
• Press Shift+Enter, or select the down arrow button, to go to a newer match.
• Press Esc to close the bar.

If a match is in older history, d5s loads it and scrolls to it. If the match is inside a closed agent message row, the row opens. Press the shortcut again while the find field is selected to use your browser's find instead.

Manage a chat

Open Conversation actions from a manual chat in the project Recents list, the All sessions dialog, the Chats list, the History list, or the open-chat header. From there you can:

• Move to project to keep the transcript but attach future work to another project's context. If this action says the chat has a coding environment, keep it in the current project until environment transfer is supported;
• Rename to change the conversation title; or
• Delete to permanently remove the conversation after confirming. Deletion cannot be undone.

In d5s Desktop, you can also right-click a conversation in the sidebar Recents list to open the same actions. Double-click a chat in that list to rename it.

d5s names a new chat automatically after the agent starts its reply. If no name can be made, the chat takes the start of your first message. A name you set with Rename always wins and is never overwritten.

These conversation actions apply to manual chats. Manage automation runs, dashboard runs, and agent conversations through their parent automation, dashboard, or agent instead.

Adjust how chats look

Open Settings → Preferences → Appearance to change how conversations render for you. Text size (Small, Medium, Large) scales the transcript text only. Width (Narrow, Medium, Wide) caps the transcript and composer columns on wide screens; on a phone the column already fills the screen. Both settings apply at once, follow you to every workspace and device, and change nothing for other members. Theme (System, Light, Dark) is also here and stays per device.

Move stable patterns onward

Once the role or trigger becomes repeatable, the same project context can support a persistent agent, scheduled automation, or refreshable dashboard. Moving to another mode reuses the project; it does not turn the chat transcript itself into standing instructions unless you explicitly carry the relevant material forward.

Google workspace protection

Google content can be used for the work you authorize in this workspace. Tasks using AI may send relevant content to the AI providers used by the workspace. Disconnecting does not erase information already retrieved. See Google data protection for AI processing and stored-data guidance.

Start a security test

If your chat offers Pentest VM, ask the agent to write and run scripts for your authorized test. Code, command output, and files use the ordinary sandbox view. The environment label identifies the Pentest VM.

Workspace owners and admins can save optional Targets and notes in Settings → Infrastructure access → Security testing. This text guides the agent. It does not restrict network access.

To supply credentials to scripts, use Secrets & variables in the agent or project settings. See sandbox credentials.

The VM belongs to the conversation. Ask the agent to export its report before stopping it or letting it expire. Its temporary disk does not preserve reports after cleanup.

A running command may return a job identifier before it finishes. Ask the agent to read that command's result. Do not repeat a command whose outcome is unknown. A successful command alone does not prove test coverage; review the agent's report and coverage gaps.

Connect AWS from a conversation

Name the AWS account and Region when asking for a spending or infrastructure investigation. A Connect AWS account card lets a workspace admin complete role setup over the chat. If access is missing, Review permissions identifies the affected operation without discarding successful results. See AWS account access.