Troubleshooting

Start by identifying the boundary that failed. Most d5s problems fall into one of five layers: access to the resource, admission of a turn, schedule eligibility, capability authorization, or execution after a run started.

Slack is connected but my first reply is missing

Workspace connection alone does not add d5s to your channel or link every teammate’s account. In Slack, open the channel details, choose Agents & apps, and add d5s. Mention @d5s with your request. If a private Connect d5s prompt appears, sign in and finish linking before returning to Slack. Use the private Choose agent dropdown if the thread has no selection; only agents you can access appear. Ask a workspace admin if your agent is unavailable. The illustrated Slack guide remains available under Manage agent → Channels → Slack, with Open Slack below it.

First-login setup did not finish

If workspace setup or saving your welcome preferences fails, use the retry action on the welcome screen. Retrying reuses an already created personal workspace and does not rename it. If a region is unavailable for new workspaces, choose an available region. An invitation to an existing workspace does not require you to create a personal workspace.

Agent setup stopped at a connection step

Selecting connectors records which services the role uses; it does not connect an account. Choose an account, connect one as a workspace administrator, or select Set up later. A deferred connector is excluded from the agent’s access until you configure it later.

After Slack or mailbox authorization, return to the same setup draft. The corresponding channel settings reopen with the result or error; cancellation preserves the draft so you can retry. An inbox marked Enabled or a connected Slack, Telegram, or feature-gated Discord identity does not mean the draft can run. Review the draft and select Activate agent before testing messages.

If activation reports a conflicting connector authentication method, change the selected connections and retry. You do not need to create another draft or channel identity. Your setup history remains available after activation in Agent settings → Setup history.

An agent cannot find a Slack attachment

Mention the agent in the same thread as the file and ask it to inspect the attachment. The agent can discover files in earlier thread messages, but thread history and attachment counts are bounded. If the context says content was omitted, attach the needed file directly to your new request. Files that Slack no longer makes available, or that exceed the supported size limit, cannot be downloaded.

No-training work is blocked before or during a run

First check Settings → Account → Privacy. If Use no-training routes for my work is on, your interactive and owned background work requires the no-training route. Also check the organization's Privacy page with an Owner or Admin: Require no-training routes for everyone applies even when your personal switch is off.

A Slack message you send to an agent is a third case. It carries your message and the agent owner's instructions and context in the same model call, so it uses the stricter of your selection and the owner's. Your turn can therefore require the no-training route even when your own switch is off. The agent replies in the thread when that routing is unavailable, and it clears its own status. Ask the agent's owner to check their Privacy page.

If no-training work cannot start or reports no eligible provider, keep the privacy requirement in place and contact d5s with the selected model and visible error. Retrying cannot bypass the requirement. Turning off your personal setting returns to normal routing only when the organization does not require no-training.

I cannot sign in

Open d5s and continue directly on the hosted authentication screen. It offers the sign-in methods enabled for d5s, including Google, GitHub, and email. Complete any email-verification, multi-factor authentication, or organization-selection step there before returning to d5s.

Public desktop releases currently open EU automatically, including when an older release saved US as its preferred region.

In d5s Desktop, sign-in uses your default system browser so you can use the accounts and profiles already signed in there. Desktop first shows Sign in to d5s and does not open another application until you choose Continue in browser. Keep d5s Desktop running, finish the browser flow, and allow the browser to open d5s when prompted. The browser then shows Browser sign-in complete and normally returns to the app automatically; choose Open d5s if it does not. You can close that browser tab after the app opens. While it waits, Desktop shows Finish signing in in your browser; choose Open browser again there if the sign-in tab was closed or lost. Desktop reopens a live attempt or starts a fresh one automatically if the old attempt has expired. If the app still does not return, quit and reopen d5s Desktop before starting a new sign-in; an old or already-used return link cannot resume a new attempt.

If d5s returns to the login page, follow the specific message shown there. Verification, MFA, additional identity checks, and organization selection must be completed in the hosted authentication flow, so start sign-in again and finish the named step. If the message says your d5s account is inactive, contact your workspace administrator instead of retrying.

On a shared device, sign out of the identity provider separately before signing in to d5s as another account. The d5s signed-out screen confirms only that the d5s session ended.

If a challenge repeatedly returns to the login page, send support the approximate time of the attempt. d5s support will never ask you to paste a verification code, pending authentication token, or OAuth token into an issue.

A run will not start

Check the visible session state first.

• If a turn is queued, the session already owns active work. Wait, or cancel the queued turn.
• If the session is active, stop it only when you intend to interrupt the current tool work.
• If an agent is paused or its daily budget is spent, manual and autonomous work can be blocked until resumed or until the next local day.
• If an automation already has a run in flight, another occurrence is skipped instead of overlapping it.
• If the response reports a credit block, the box you sent from replaces its send button with a message that names the cause: an empty balance, a full pace window, the member's spend limit, or no active plan. It gives the time work resumes when one is known and presents available billing actions to an organization Owner or Admin. Read that message as the authority on whether a turn can start. See Plans and credits.
• If the request is rejected before a session appears, verify workspace access, project access, model policy, and uploaded file ids.
• If a newly listed model is missing from the composer or cannot be selected, ask a workspace Owner or Admin to enable it under Workspace settings → Models. New catalog entries are not enabled automatically.
• If every model is unavailable in a new organization, ask a workspace Owner or Admin to confirm that GPT-5.6 Luna is enabled. Free organizations can use only Luna; other models require a paid plan and must be enabled for the workspace. Contact d5s with the organization name if an expected model is unavailable.

If a workspace is missing from the switcher and an Other regions section is shown, use Open region to check the region named in your invitation or onboarding instructions. If that destination says Coming soon, it is not available and has no link. If a direct link instead says Workspace isn't in this region, use Try region or Choose another region. The requested page path is preserved, but query parameters and fragments are removed and you may need to sign in to the destination.

d5s deliberately does not claim that a neutral region link contains your organizations. The destination asks you to choose an account on its hosted sign-in screen; select the account named in the invitation. After the regional app opens, look for the organization and workspace there. If either remains missing, ask an organization Owner or Admin to assign you or one of your groups from the organization's Workspaces page. An organization Owner or Admin, or an Owner or Admin of the workspace itself, can instead invite you by email from Settings → Organization · organization name → Members → Invite people; the workspace's own Members section grants access to people who are already in the organization, and opening that section needs an organization role even for a workspace Owner or Admin. Organization membership and a workspace grant are separate; joining the organization does not grant access to all of its workspaces.

If Usage → Workspace stats shows a Historical workspace label, the usage entry still refers to a workspace ID whose current name is unavailable. This preserves the organization's totals and does not indicate that the workspace can still be opened. The table covers the complete selected period; its chart says when it is limited to the most recent 366 days.

Understand what an agent saw during a billed run

An organization Owner or Admin can open Usage → Activity, select an available row, and choose Inspect run. Review the billed outcome, recorded activity, model usage, and any failure shown. Use Open conversation to jump to the related turn when it still exists.

A run record may not contain the exact input sent to the model provider. Missing evidence is not a reconstruction of that input. Billing access does not grant access to conversation content, and a record is unavailable when the conversation has been deleted or you no longer have permission to view it.

A sandbox id can be empty before the run needs an execution environment. That alone is not a failure; look for the queued, start, sandbox-ready, error, or terminal events.

A primary model response can continue through periods without new output, within the run's overall time limit. Initial response recovery, context preparation, and tool execution have separate limits. Before retrying a failed run, check its recorded activity for completed external actions so you do not repeat them unintentionally.

A message send lost its response

Open the same conversation and inspect its recent history before sending a new copy. A connection failure can happen after d5s accepted the message, and acceptance does not mean the agent has finished.

For an API integration that assigned a client request ID to a supported persistent-agent send, retry with the same user, workspace, session, ID, and exact original payload. Do not edit the payload or generate a fresh ID just because the first response was lost. Follow the recovered turn in history. If the server reports a conflict or rejects keyed sends for that session, inspect the request and use the developer retry guidance before trying again. This contract does not change the separate retry rules for schedules or external message delivery.

A connector action is unavailable

Check the connector chain in order:

1. The workspace credential exists and is not archived or awaiting reauthorization.
2. The credential has not been removed from the session's project. For an agent, inspect the agent's home-project connectors.
3. The requested tool is Allow, an Ask approval is still live, or the session is visibly in Allow all mode.
4. The requested action is within the connector's supported access.
5. The external account can access the requested object.
6. The provider account has sufficient credits, quota, and feature access for the requested operation.

When several connections use the same provider, confirm that the intended named connection remains attached and that its stable handle is the one the agent used. If an Attach card says the connection is no longer available, refresh the conversation and search again; the credential may have been archived, moved into reauthorization, or already attached since the suggestion was created.

If an Ask action is waiting, return to the conversation and answer the pinned connector approval card. Approving once does not approve later calls. You can instead switch that conversation to Allow all, which retries the pending action once. If the retry fails, the approval card remains available. If Allow all is not shown or an action still pauses, the feature may be unavailable for the workspace or an explicit Deny or another connector boundary may be blocking the call.

The provider changed its tools

If the agent reports that a tool definition changed, have it inspect the same connected account again and check the current operation arguments. Do not switch accounts to work around the error. A discovery error is different from a successful empty tool list. An administrator can refresh the credential's reviewed tool inventory; a failed refresh keeps the previous snapshot.

If an operation times out or its output cannot be saved, it may already have executed at the provider. Check the provider's state before repeating a write. A returned resource link is a reference, not proof that its target has been downloaded.

If a connected account remains unavailable, check its authorization and reconnect it if requested. If the error persists, contact support with the connection name and visible error.

If a remote MCP connection fails before the provider's sign-in or consent page appears, retry once and then contact support with the provider name, approximate time, and visible error. The provider may be unable to read d5s's public client metadata or may not accept its client domain. Never include an OAuth code or token in the report.

A scheduled run did not happen

For an automation, verify enabled state, trigger kind, timezone, next due time, daily quota, and whether a run is already active. Missed recurring slots coalesce into one due run. A one-time occurrence retries until accepted, then disables itself.

For an agent scheduled wakeup, verify its enabled state, the agent pause state, timezone, next run, last status, and daily budget. A busy or transiently failed occurrence is retained for retry. A one-time wakeup also remains due after a budget skip and retries when budget becomes available.

Remember that Wake now does not move an agent's scheduled wakeups and Run now does not reset an automation's recurring schedule.

An agent did not enter a scheduled call

Open the agent's Calls → Auto-join tab. Confirm that the intended calendar is selected, the joining rule covers the event, the event is accepted when the rule requires that, and its individual switch is on. If the meeting came from another calendar, add the agent address shown under Invite from any calendar as a guest; the event must also appear on a selected calendar.

Then open the meeting under History. Waiting for host means the listener is still in the lobby. Not admitted means the host did not admit it before the join window ended, and Declined by host means the request was refused; ask the organizer to admit the listener or include its address on the invitation before retrying. Call failed is a less-specific provider failure that can happen before or after entry, so verify that the meeting link still opens and that the meeting service is available. A transcript cannot begin until the listener enters the call.

An automation completed but its agent is still waiting

Check the execution status and each agent's handover status separately in Latest run or History. Pending handover means that agent has not received the result through an eligible conversation yet. Received confirms receipt, not human review or follow-up work. A successful run can still be waiting for an agent, and a failed run can have a received handover.

Select Open run or Open to inspect the complete result directly. If the handover is unavailable, use Retry; do not treat a loading failure as confirmation. People with access can inspect and review the run without an attached agent.

Agent email did not send

Open the agent conversation and inspect the outgoing email card first. Under Manage channels → Email, select the sending mailbox and check its Outgoing rules. Confirm that the agent is not paused, its AI budget and email daily limit remain available, and the recipient is not suppressed. New email needs a selected or default mailbox; replies stay with the mailbox that received the original message. New automatic d5s inboxes require approval before sending. Existing policies may also require approval or restrict the agent to replies until explicitly changed.

An email activity state of Sent means its mail provider accepted the request; it does not prove recipient delivery. A timeout after submission is left Uncertain instead of being retried automatically, because a blind retry could send a duplicate. Failed and blocked entries include the reason recorded by d5s.

If incoming mail does not start agent work, check that the receiving mailbox is active and its Incoming rules allow the sender. Under Advanced → Blocked senders, a matching block overrides all incoming permissions. When Accept replies from people this mailbox emails is on, an address this exact mailbox successfully sent to can also email the agent; drafts, failed or uncertain sends, and another mailbox's history do not qualify. This exception does not grant outgoing permission. When Review unknown senders first is enabled, successful-send history still does not bypass review; only entries shown under Trusted senders do.

For Gmail, confirm the message reached the connected inbox and reconnect if mailbox authorization has expired. Allow time for the next mailbox check. Automatic responses and common bulk or mailing-list messages start ordinary agent work only when their From address explicitly matches the mailbox's saved Incoming rules; the successful-send reply exception is not enough. For a d5s address, inspect Email activity for domain, spam, virus, or other incoming-mail check failures. An explicitly allowed message may use a different Reply-To, but replies remain addressed to its From contact. Read Agent email for setup and address-rule details.

If an incoming email attachment is not read, inspect its row on the email card. Unsupported or unsafe files are skipped with a reason. Supported files are fetched only when the agent needs them, so a reference can remain unread without indicating a failure. If a previously readable file now fails, confirm that the original message and attachment still exist in the same connected mailbox and reconnect that mailbox if its authorization has expired. For a d5s address, an unread attachment can become unavailable after the original message's 30-day retention period; supported files already downloaded remain under the agent's retained file controls.

Slack did not ask which agent should answer

Invite @d5s to the channel and select it from Slack's mention suggestions when sending your request. If a thread already has an agent, or the channel has a default, d5s uses that choice without asking again. Use Change agent in the thread to choose someone else. A thread choice takes priority over the channel default.

If d5s asks you to connect your account, use the private Connect d5s action with an account that can access the agent. Only the requester can use their agent picker. If the picker expires, mention @d5s again. If an agent is missing, confirm that you can open it in d5s and that it is available for use in this channel.

The optional shortcut is @d5s +poppy investigate, with no space between + and the agent name. It does not autocomplete. Unknown or ambiguous names are refused; use the picker instead of guessing a handle.

Shared Slack replies fail or still look like @d5s

The reply body shows the selected agent's name and character so the identity remains clear when Slack groups messages. The native Agent name · d5s sender also needs Slack's customized-message permission. For a connection installed before that permission was added, a workspace admin must open Manage agent → Channels → Slack and choose the Reconnect Slack link under Open Slack once before customized sends can resume. With several connected Slack workspaces, the sentence names each workspace as its own link.

A Slack coworker ignores replies in a thread it follows

A coworker installed before thread follow shipped still runs its older Slack app configuration, so Slack does not send un-mentioned thread replies to it. Such a coworker does not promise to follow: its first reply in the thread says it cannot follow the thread yet and links to Upgrade permissions in d5s. Follow that link, or open Manage agent → Channels → Slack → Dedicated identity, where a line under the identity names the pending upgrade. A workspace admin chooses the highlighted Upgrade permissions button, continues through Slack OAuth, and approves the added permissions; the existing app, channel connections, and history are retained. Threads that received the prompt are not followed retroactively; mention the coworker once more in each of them. Then confirm Thread follow is on for that channel, that the thread was quiet for less than 72 hours, and that nobody reacted to one of the agent's messages in the thread with :no_bell:. A fresh mention in the thread starts the follow again.

A Slack coworker message did not arrive

If the text message itself is missing, inspect the outgoing activity before asking the agent to send it again. Accepted by Slack means Slack accepted the API request; it does not prove teammate receipt or a read. Unknown means the request may have crossed Slack's acceptance boundary, so d5s avoids a blind retry that could create a duplicate. Check the target Slack channel or thread for the original message first.

An Unknown entry can also say that Slack explicitly refused the first attempt before acceptance and d5s queued a bounded retry. That retry rechecks the current agent, originating user or channel grant, channel binding, and Slack credential. Pausing the agent, disconnecting the channel, revoking access, or removing the Slack installation stops the delayed send. Do not submit a fresh copy while that retry is pending.

A dedicated Teams agent is not ready or does not reply

Open Manage agent → Channels → Microsoft Teams. Where this setup is available, an organization Owner or Admin with workspace administration access can create and manage the identity. Installation required means the identity is saved but not ready in a team: select Continue in Teams, complete Microsoft installation, mention the named agent in an intended standard channel, then refresh and activate that verified destination in d5s.

For thread replies without another mention, approve Upgrade permissions in Teams, return to d5s, select Refresh permissions, and enable Follow replies for that channel. A grant for another team is not enough. If provisioning needs reconnection, reconnect it before retrying the saved identity. Renew credentials replaces an expiring credential without asking you to copy a secret.

A failed management action can leave an existing connection available; retry the failed action instead of creating a duplicate identity. Cleanup pending means the identity has stopped in d5s but Microsoft cleanup is unfinished; select Retry cleanup. See Teams agent setup for the supported scope. A response in d5s is not proof that a message was sent to Teams.

A Telegram teammate bot is missing or does not reply

If the bot is not connected, an Owner or Admin can select Add Telegram in the agent’s Channels panel and follow the illustrated setup guide. Bot connected does not mean your personal Telegram account is linked: open the bot, send /start, and confirm the d5s link before testing a reply. If the token is rejected, the setup form below the guide shows the error and lets you correct it. Use Previous and Next to revisit the instructions without clearing your token or the error, then try connecting again.

Telegram teammate DMs are a limited beta. If Manage agent → Channels → Telegram is missing, confirm that d5s has enabled the beta for your workspace. Connecting or managing a bot also requires a workspace Owner or Admin who can share and run that agent.

The agent conversation’s Channels panel shows Telegram status and Open bot when connected. Could not load Telegram means the status request failed; use Retry rather than treating the bot as disconnected. For Needs attention, ask an authorized Owner or Admin to open Manage and restore the connection.

If a bot will not connect, create or select it in BotFather, copy its current token again, and confirm that the bot is not already connected to another d5s agent. Each named agent needs a different bot. Never send the token in chat or a support request.

If the bot asks you to link, use a private chat with the bot, send /start, and open the returned link in a browser where you can sign in to d5s. Confirm the displayed Telegram identity with an active d5s account that still belongs to the workspace and can view and run the agent. The link expires after ten minutes; after it expires, send /start again for a fresh one. Repeated messages during the pending ceremony do not produce another link. A group member, bot account, unlinked account, or teammate whose current d5s access was removed cannot start work.

The beta accepts private text messages only. Groups, channels, files, buttons, edited messages, and other non-text updates do not wake the agent. Replies do not stream, and the agent cannot initiate a Telegram conversation, so inspect the explicit delivery activity and the same private chat. An internal completion message alone does not mean a reply was sent.

Telegram messages use a Telegram badge. A sender may still appear as initials while their profile photo loads or if it is unavailable. Photo loading does not delay incoming messages. Disconnecting Telegram can make these photos unavailable.

If the outgoing activity is Accepted by Telegram, Telegram accepted the API request but has not proven teammate delivery or a read. If it is Delivery unknown, check the private chat for the original message before retrying; d5s will not blindly repeat an ambiguous send. Disconnecting the bot removes its d5s links and bindings but does not revoke its BotFather token. Revoke or rotate that token in BotFather if it must stop working. To connect again, paste a token for the same bot or for a new bot; the agent keeps one Telegram bot at a time, so disconnect the current bot before you replace it.

A Slack coworker did not receive or send a file

For an incoming file, first confirm the message shows an attachment reference. Files are not copied into the sandbox automatically: the agent uses its native download tool when the task needs one. Then open Manage agent → Channels → Slack → Dedicated identity. If the installation shows Upgrade permissions, continue through Slack OAuth to add files:read; older coworkers do not gain that scope automatically. If d5s says the provisioning connection is unavailable, a workspace admin can reconnect it in that same agent panel and d5s retries the upgrade automatically. Also confirm the coworker is still installed in the channel and Slack still retains the file. Files above the per-file limit or beyond the per-message count are skipped with a note in the turn.

For an outgoing file, ask the agent to attach a file it created and confirm that it is available and within the attachment limits. Missing, empty, oversized, or unsupported files are refused, as are attachment sets above the count or total-size limits. The text reply can still arrive; inspect its attachment-failure note for the reason.

Downloaded channel files normally remain available after the coding environment stops. If a previously downloaded file is missing, give support the conversation link, message time, and visible file error. A file that was never downloaded still depends on the original remaining available in Slack.

An invitation is refused for lack of seats

For teams that purchase seats upfront, invitations cannot exceed purchased Team capacity. Active members plus pending invitations must stay within the effective seat quantity, and both organization invitations and workspace invitations count. Raise capacity from seat management or withdraw a pending invitation before trying again.

Two cases look like a full team but are not. A team organization that has never completed checkout has no Team capacity at all. Open its Settings → Organization → Billing page, select Change plan, then choose Choose seats and start Team to finish the subscription on that organization. When a paid seat decrease is already scheduled for the end of the period, new invitations are measured against the lower upcoming quantity, not today's; cancel the decrease by requesting the current quantities again if you need those seats back. Personal organizations are unaffected: their shared-workspace guests are never billable seats.

A seats-on-joining invitation needs review

If your team uses seats on joining, select Review invite and have an organization Owner or Admin authorize the displayed joining charge or free offer and individual monthly price. Quotes last up to 15 minutes. A free invitation reservation lasts up to 48 hours, bounded by the team's first billing-period end. If either expires, review and authorize new terms; the expired free offer does not silently become a paid charge.

A pending invitation is uncharged. If payment fails or billing terms change before acceptance, ask an organization Owner or Admin to authorize a new invitation. If acceptance says the purchase is being reconciled, retry as directed. Contact support if it remains blocked or requires billing reconciliation. Do not send repeated replacement invitations to bypass that state. Someone already in the team needs no additional seat and should open the workspace named in the invitation.

A seat change did not affect credits

Check the billing summary's standard and premium seat quantities and the effective date shown for the change. Increases apply after successful payment, while removals and class downgrades take effect at the end of the current billing period. If the displayed credit balance still looks wrong after the change becomes effective, contact d5s with the organization and billing period. See Plans and credits.

The result looks stale

• A dashboard may be showing the last live version while a new build is active or failed. Open the chat and version menu to inspect both.
• A project file or connector activated after a run changes future runs, not the history of the completed one.
• An agent keeps one long conversation and may compact older context; put durable facts in project instructions, files, or memory rather than relying on a distant chat turn.
• A restored dashboard version becomes live without creating a new agent run.

I cannot find a workspace setting

Type what you are looking for into Search settings at the top of the sidebar. Matching workspace sections are listed by name under the Workspace · workspace name group, and each one opens that section directly, so you do not have to know which group holds it. The search matches related terms as well as section names — egress finds Network, and invoice finds Billing. Clear the box to return to the ordinary sidebar.

Without a search, the sidebar lists account and organization administration, then a Workspace · workspace name group whose Workspace settings entry opens the sections for the selected workspace. That entry stays highlighted while you are in any of them, so it is also the way back after visiting an organization page. Choose a section from the row across the top of the page: General is available to workspace members; Members lists who can open this workspace and is administered by an organization Owner or Admin; Models, Network, and API Keys require a workspace Owner or Admin. On a narrow screen, return to the Settings index first; workspace settings appear as their own section below account and organization administration.

A section your account cannot administer is absent from the search results exactly as it is from the rest of the sidebar; searching never reveals a page you could not already open. Which role a section needs varies, so check that before asking: Models, Network, and API Keys need a workspace Owner or Admin, while the workspace Members section is administered by an organization Owner or Admin — a workspace Owner or Admin who is an ordinary member of the organization does not see it. The organization's own Members roster is a different section under the Organization heading, and every member of the organization can open it.

If you can open General but cannot edit Workspace name, your workspace role does not include administration. Ask a workspace Owner or Admin to rename it from Workspace settings → General → Workspace details. An organization Owner or Admin without access to that workspace cannot rename it from the organization portfolio. Renaming changes only the display name, so the slug and existing links remain the same.

To rename the organization itself, use Organization > General > Organization details. That page is available only to organization Owners and Admins. If it is missing, confirm that the Settings scope tag says Organization and that your organization role is Owner or Admin. An organization rename also leaves the slug and existing links unchanged.

A workspace cannot be deleted

Only an explicit workspace Owner or Admin can open the deletion readiness check, and only the current Owner can request the email code and schedule deletion. Organization administration alone is not enough. Personal workspaces cannot be deleted.

For another workspace, open Workspace settings → General → Danger zone, select Check readiness, and follow each Review link until every count reaches zero. A pending provider revocation must finish too. After the Owner confirms the exact workspace name and email code, deletion remains cancellable for 48 hours. If a dependency returns or ownership changes during that period, d5s automatically cancels deletion and leaves the workspace intact.

If the workspace has disappeared but the completion email has not arrived, deletion may still be finishing or the email may be delayed. Do not recreate the workspace to force progress. Contact support if the email remains absent.

d5s Desktop is not updating

Settings → Desktop → Updates shows the installed version, a manual check button, and a live status line. Automatic updates download in the background and are installed when you choose Update now in Settings, select the blue download icon beside your avatar in the expanded sidebar, or quit the app. Before installing, d5s checks for a newer release and replaces an older downloaded update, so you can skip intermediate versions. Update now reopens d5s; a normal quit leaves it closed.

• macOS: "Automatic updates are unavailable in this build." This build cannot verify update signatures and cannot install an update even if one is downloaded. Download the latest installer from d5s.tech instead of waiting for it to update itself, and repeat that until this message stops appearing.
• A background check runs at launch and again every few hours; it fails silently so a temporary network problem never interrupts you with an error. Only a manual check (the button in Settings, or "Check for Updates…" in the app menu) surfaces a failure.
• Once a download finishes, a small blue update button appears beside your avatar in the expanded desktop sidebar. Select it to install and reopen d5s. When the sidebar is collapsed, a small blue dot at the avatar's upper-right indicates the ready update but is not a button. Open Settings → Desktop → Updates to install, or quit the app. The indicator is absent while an update is still downloading and on pages without an avatar, such as sign-in. If a newer release becomes available, d5s downloads that version instead.
• If downloading a newer replacement fails in the background, Settings shows an error so you can retry; it does not report that the app is up to date.
• If d5s cannot check the latest release or finish downloading it, Update now reports a failure so you can retry. If an avatar-side install fails while that update remains ready, the expanded button shows "Update failed. Try again." and re-enables. If a newer replacement is downloading or fails, check the status in Settings and retry there. A normal quit still closes the app without installing the older queued version.

Report a bug

Select Report a bug in the sidebar, describe what went wrong, then select Send report. The report is private to the d5s team. Your account, workspace, current conversation when available, and available technical details are included automatically; you do not need to look up identifiers or device details.

A screenshot is optional. In a desktop version that supports capture, the app adds a preview of its visible content before the dialog opens. Review it and select Remove if you do not want to include it. In the browser or an older desktop version, upload or paste a screenshot instead. You can send a report without an image.

If sending fails, your description and screenshot stay in the dialog. Select Try again. After Report received, select Done to return to your work. The team can follow up at your account email address. During beta, the feature request board is hidden from the sidebar. Existing direct links to the board still work.

If you cannot open the app, contact support with a short description and the approximate time of the problem. Never include API keys, OAuth tokens, connector secrets, or confidential prompts.

An agent says a mailbox-channel email is still awaiting approval

Ask the agent to check the email's current outbox status. It can inspect approval and delivery records without sending another copy. A previous “queued” reply is historical; a later approval or send may already have completed. If the outcome is uncertain, avoid retrying automatically because the provider may already have accepted the message.

Agent files have not reached my Mac

Check Manage agent → Local files → Recent operations and the connection status. Keep d5s running on an awake, connected Mac. Resume paused access in Desktop settings. For an unavailable folder, select it again and review permissions. A read requires Read only or Both; a write requires Write only or Both. For an unknown write outcome, check the destination file before repeating it. Files presented in the conversation are no longer saved automatically. See Local files.