What brand chat can do
- Write and revise on-brand copy.
- Work on your own website, other first-party sites, social profiles, and key pages without being given the URL. daoco carries them in brand context from the sources you used during onboarding, so “look at our pricing page” is enough.
- Generate, edit, and resize visuals.
- Create draft social posts and prepare publish approvals.
- Run research, SEO, campaign, clipping, and analytics tasks.
- Open workflow panels for richer review and editing.
- Refresh brand context after you approve the proposed changes.
- Work with connected integrations when the workspace has them enabled.
- Copy structured text blocks, such as hooks or campaign copy, directly from assistant responses when daoco renders them as copy cards.
- Pause, resume, stop, retry, or queue follow-up instructions while longer work is running.
- Follow a live, turn-by-turn execution ledger and stop an individual delegated branch when the control is available.
- Send a result to yourself through your assistant on WhatsApp, Slack, or Discord, or ask to be pinged there when the work finishes.
How chat routes your request
daoco uses one main chat thread for most work. When you send a message, daoco decides the best specialist for the job:- Slash commands. A command such as
/postor/researchstarts with that specialist. See Assistants and slash commands. - Follow-up in the same workflow. A continued post, clip job, campaign, or brand refresh stays with the active workflow instead of restarting.
- Natural language. For everything else, daoco interprets the request and may run one specialist or a short sequence, such as research before writing. Mentioning social tools in a research question or asking for a short comparison does not request social-post or video work.
Run specialists at the same time
daoco can hand parts of a request to several specialists at once, for example researching competitors while drafting a post. Adjacent specialists sit under one header. While any of them is still going it reads “Kicked off 3 specialists · 2 working”, and once they all finish it reads “Ran 3 specialists · ✓ completed”, or names what didn’t finish (“1 failed”, “1 stopped”). A specialist is only counted as completed when it actually completed. The Parallel toggle in the composer, next to the approvals shield, sets this per thread. It is on by default, and specialists can work at the same time. When it is off, daoco runs one specialist at a time. You can set it before your first message in a new chat and it applies to the thread that message creates. Specialists that can ask for your approval, such as the ones that publish, schedule, or change a connected app, always run alone, even with the toggle on.The specialist view
Click a specialist’s row to open its view in the chat column. The view looks like the thread: the specialist’s own transcript (its tool calls, its notes, any specialists it handed work to, and its result) in the same column, with a bar where the composer usually sits. If the specialist produced something for the side panel, such as a draft or a report, the side panel shows that specialist’s items while its view is open, and the thread’s panel comes back when you return to the thread. The view’s address includes the specialist, so refreshing the page or sharing the link opens the same view. The view is read-only. The bar shows the specialist’s type and title, its status, how long it has run (frozen once it ends), and that it runs on its own. Stop ends that one specialist and leaves the rest of the run going; the specialist then reads Stopped, not failed. A specialist waiting on your approval has no Stop button: answer its approval in the thread instead. Back to thread or Escape returns to the thread with its scroll position and your draft intact.Approvals with several specialists
Approvals are always asked in the main thread, never inside a specialist’s view. When more than one specialist is waiting on you, each approval gets its own card above the composer, oldest first, and you answer each one on its own. A card raised by a specialist says which one is asking, for example “Research asks”; click the name to open that specialist. When a specialist’s view is open and it is waiting on you, the bar says Needs your approval and Review in thread takes you back to its card. If you type “approve” instead of clicking, it answers the approval only when exactly one is waiting. With more than one, none is answered and daoco asks you to use the cards. Bypass approvals and your approval preferences apply to every specialist in the thread the same way. In the sidebar, a thread with more than one specialist working shows how many are running under its title.Control active runs
Long-running work can involve research, publishing checks, campaign planning, or specialist tasks. When a run is active, the composer can show controls for the current state:
Some pauses are waiting for a required action, such as an approval, reconnect, or browser auth. Complete the card shown in chat instead of using Resume for those pauses.
Attach images
You can attach up to 50 images to one message. Multiple images appear in a thumbnail grid; click a thumbnail to open the gallery and use the arrow keys to move between images. A follow-up without new files keeps the most recent upload available for up to 24 hours, including the whole set of images.Send follow-ups while work is running
You can send another instruction before the current run finishes. The follow-up appears immediately at the bottom of the transcript as a faded, dashed bubble marked Queued, under the line that shows what the current run is doing. That line always belongs to the work above it, never to your queued message. Each queued bubble carries its own controls: Steer, Redirect (when the run can be interrupted), Edit, the up and down arrows, and Cancel. Steer and Redirect only appear while the run can take a correction; when it can’t, for example while a specialist is still working or the thread is being compacted, the bubble says so and when Steer comes back. Edit changes the text inside the bubble. The message keeps its files and its place in the queue, and whatever you were typing in the composer stays where it is. Press Enter or Save to keep the change, or Escape to discard it. If the run picks the message up before you save, the original text runs, and your edited text goes to the composer so you can send it next. The arrows move a message up or down the queue, which changes the order the messages run in. Cancel removes a message before it reaches the model. Long messages collapse to a few lines with a Show more toggle, and attached files show as thumbnails. When the run picks the message up, the bubble becomes a normal message just above the working line. While that happens it reads Sending… and can no longer be cancelled. Each queued message says when it will run. An ordinary follow-up says Runs when this response finishes and waits for the whole answer. A message you steer into the run (click Steer, or press Cmd+Shift+Enter / Ctrl+Shift+Enter to steer the oldest queued message) says Folds in after the current step: the run picks it up at the next break between steps, usually within a step or two rather than at the end. When a steered message lands, the transcript marks the spot with a Steered divider, so you can see which work came after your correction. Files on a steered message reach the run with it and show under the divider. Corrections stay with the active run across retries and continuation steps. Your latest correction can change the original plan, while approval and publishing restrictions still apply. Steering waits for the step the run is in to finish. When you would rather not wait, a queued message also offers Redirect: the run stops the step it is in, folds your message in immediately, and carries on as the same run. Because it stops work in progress, the bubble asks you to confirm first with Redirect now and names what it will throw away — the tool call it cancels, and anything the current step had written. Finished steps are kept, and the transcript marks the spot with a Redirected divider. If you click Stop run while the redirect is taking effect, the run stops instead of starting another step. If the run ends before it absorbs the message, the bubble clears the Redirecting label; a later run does not inherit the redirect. If the run cannot be interrupted right then — a specialist is running, it is waiting on an approval, or it has not started work yet — the message stays a steer and the bubble says so. When the active run is safe to steer, a plain-text follow-up can guide its next step without waiting for a second turn. If the run is compacting, paused for approval, or otherwise stops being steerable before the message lands, the bubble says Couldn’t fold in; will run when this response finishes and it runs as an ordinary follow-up. Compatible plain-text messages can be combined into one follow-up turn; messages with attachments stay separate. If a follow-up cannot start because the workspace is out of tokens, it waits in the queue marked Out of tokens with your text and files intact, and runs on its own once the balance is back. Cancel it if you would rather not wait. If your workspace already has as many runs going at once as its plan allows, a new message waits in the queue as well, and starts on its own as soon as one of those runs finishes. Use Stop run when you want the current run to end now. The transcript reads Stopping… until the backend finishes checking the run, then You stopped this run. Stopping closes any approval the run was waiting on, so an old approval card can’t restart the stopped work. Messages you queued behind it are held, not sent: each bubble reads Held after you stopped and offers Send now, Edit, and Cancel. Send now runs it as the next message, or queues it behind a run that is already going; nothing held runs until you choose. A new message you send while others are held starts on its own and leaves them held. Messages that came from your team assistant or a connected AI app stay queued and run next, so the person who asked still gets an answer. If you paused the run yourself, queued follow-ups wait until you resume, and their bubbles say so.What the loading line is telling you
While a run is working, there is always one line under the transcript. It names what the run is doing now: understanding the request, planning, drafting, working through the steps, handing off to a named specialist, reviewing what that specialist found, writing, or continuing when a long turn picks back up in a fresh step. Beside it are the tool in use and how long the turn has been going. The line stays on its stage for as long as that stage is true, and only says “Working on your request” when nothing more specific is known. Between a finished run and a queued follow-up starting, it reads “Starting your follow-up”. When a specialist’s row is already showing its own progress, or daoco’s reply is streaming in, the line shrinks to just “Working” and the elapsed time, so you never see two spinners and never see none. Once the answer is written and the run is only wrapping up, it reads “Finishing up”. A follow-up you queued starts its own clock when its turn starts, not when you typed it.Interactive visuals
daoco shows charts, tables, diagrams, and comparisons as interactive visuals directly in a reply. The orchestrator and the question, research, SEO, and campaign specialists create them; writing and social specialists answer in text. Visuals can open in the workflow panel and remain available when you return to the thread. Charts in replies from before this change keep their original look. Charts built from stored brand data read the saved figures for published posts, content analysis, keyword ranks, and AI visibility. Hover or focus the visual to download it as HTML or PNG. Reports sent through an assistant on WhatsApp, Slack, or Discord include a PNG when rendering succeeds. The page runs in a sandbox that blocks access to your dashboard and external resource requests. Browser sandboxing does not fully prevent a page from navigating itself to another URL, so it does not guarantee that data embedded in a visual stays offline.Work safely
daoco asks for confirmation before sensitive actions such as publishing, scheduling, applying brand updates, or making provider changes through connected integration tools. The approval card identifies the requested operation, target connection, and relevant scope. Review those details with any previews, diffs, attachments, and dates before approving. Approving an ordinary provider write runs exactly that action and also lets daoco keep making updates on the same provider for the rest of the thread without asking again, so one approval covers the whole job. Destructive changes, such as deleting something in a connected app, are always approved one at a time. The shield toggle in the composer turns on Bypass approvals for the current thread: gated actions, including publishing, run without asking until you turn it off. Turning it on while a connected-app write approval is already waiting answers that approval too, so the paused run resumes on its own; other kinds of waiting approvals keep their card and still need your answer. Deletes still ask, actions that always require confirmation (like a paid final render) still ask, and a connection that is turned off or read-only still refuses. The toggle is per thread, requires the workspace owner or admin role, and stays on across refreshes. Turning it off also withdraws the provider access the thread was granted while it ran open, so the thread goes back to asking. A workspace admin can turn the feature off entirely under Settings → Team → Approvals, which hides the toggle and puts already-bypassed threads back to asking. Brand owners and admins set approval preferences per brand. Ask before acting is the default. Work autonomously allows ordinary connected-app updates without asking, while Full autonomy can also publish and send. Per-connection exceptions can make access stricter or allow ordinary updates on one connection.Browser auth for social tasks
Browser automation can use a secure browser session for supported UI tasks on LinkedIn, Instagram, Facebook, X, Threads, TikTok, YouTube, and Bluesky. When required, daoco shows a browser auth card in chat. Open the secure browser, sign in to the selected platform, then confirm on the card so daoco can resume the request. You can clear the browser session from Connections without disconnecting the platform’s OAuth or API connection. See Browser automation.Chat activity signals
When a background run finishes in another thread, the sidebar can highlight that thread until you open it. This is useful for long research, scheduling, brand refresh, and generation work that may complete after you switch surfaces. Each thread in the sidebar shows its state. A paused thread says what it is waiting on: Needs approval, Reconnect needed, Sign-in needed, or Paused by you. A thread you stopped shows Stopped, not Failed or Done. A run that ended because the workspace is out of credits, or because a usage limit was reached, shows Limited with the reason, in the sidebar and in the thread’s status row. That is not an error in the work: add credits or wait for the limit to reset, then retry. Work that happens outside the thread you have open shows up under Updates in the sidebar, below Assistants. It covers assistant tasks that reported back or are waiting on you, and open items in the brand’s queue. The section stays collapsed with a count until you open it. Selecting an update takes you to its thread. Dismissing an update hides it for you only. It does not approve, cancel, or resolve anything. Updates about the thread you are viewing are left out. To hide the section, turn off Show background updates in your account settings. Updates you already dismissed stay dismissed if you turn it back on. While a turn is running, the transcript shows a single live status and elapsed time. The execution panel breaks the request into plain-language steps such as Understanding the request, Researching, or Writing the copy. A required approval changes the relevant step to Waiting for you. Finished turns keep their duration and can be reopened in the ledger. As a multi-page SEO audit grows, consecutive updates appear as one audit entry that opens the latest combined report. Separate audits and failed steps remain visible. The composer context meter shows current context pressure and, once usage has been billed, how many basic tokens the thread has charged so far. Hover it for in, out, and total thread usage. See Understand token usage. Long tasks can continue across execution windows using saved progress. Image sets are tracked by saved item, so later windows can finish missing items without replacing completed ones. A run that reaches a limit or stops making progress reports the remaining work. Review the saved outputs if the run stops early. Stopped runs show a stopped state instead of a generic failure. When the run stopped part-way through a step, the composer offers Continue run: daoco picks up from the interrupted step and reuses everything that already finished instead of starting over. Retry run is offered when there is nothing to pick up from, and starts a fresh attempt from the latest request. Edit and resend, above the composer, reruns the same request with wording you change; any files on the original message are kept. A processing-limit error means the agent run stopped; it does not by itself mean an attached file is invalid. Saved outputs remain available for the unfinished work. A resumed turn — after a reconnect, a sign-in, or a Continue — is marked in the transcript with a short “Continued” divider rather than a message that looks like you sent it. If a connected-tool step fails, daoco separates connection setup problems, expired authorization, provider refusals, and temporary transport errors when it can. A retry keeps the original connection and approval scope. Reconnect only when the recovery card says the provider authorization needs attention.Use chat with product surfaces
See Workflow panels for the right-side review surfaces and Assistants and slash commands when you want explicit routing.
Click an image attachment to enlarge it in chat. Close the preview or press Escape to return to the conversation.