Chat with Sparky

A conversation with the brand's assistant. It answers questions about the brand, proposes and runs automations, and produces content in the conversation rather than sending you elsewhere for it.

/brands/{brand}/chatAnyone with access to the brand. Editing instructions needs Editor, Manager or Owner.

What Sparky knows

Sparky reads the brand's knowledge before every reply: what discovery found, the documents you uploaded, what it has learned from your feedback, and the instructions set at organization, parent brand and brand level. It is not a general chatbot pointed at your data. It is the same brand context every automation runs against.

Instructions cascade. The organization's apply to every brand and cannot be overridden; a parent brand's add to those; the brand's own add again. A brand can add detail but never contradict the organization.

If the brand's knowledge cannot be read, a run fails rather than inventing content. Generating against constraints it never saw is worse than not generating.

The conversation

Type in the composer at the foot of the page. Enter sends, Shift and Enter make a new line, Escape clears the draft. Messages can carry up to three attachments, fifteen across a conversation, at 10 MB each for anything but video.

In the transcript	What it is
Question card	A question with answer buttons. Picking one sends it. Some accept several answers with a confirm step, and some accept free text instead, which the card says.
Automation proposal	A card naming an automation, why it is suggested, its credit cost, and a summary of its settings. Edit configuration opens the full form.
Run card	An automation working or finished, with a progress tracker and the result when it lands. Carousels show slides appearing as they are written.
Calendar proposal	Proposed posts with dates and platforms, each toggleable. A note box asks for changes in plain words, so revising is a sentence rather than a form.

A message written by an automation rather than by Sparky carries a From automation badge.

Approving an automation

When Sparky proposes running something, the composer becomes a decision bar: the automation's name, Cancel, and Run automation. You cannot type past it, which is deliberate: a proposal is a decision, not a suggestion to ignore.

Several proposals are worked through one at a time, with a note saying how many are waiting.

Sparky can run any automation the brand has. The card names the credit cost before you approve it, and nothing has to be turned on first.

If the cost changed between the proposal and your approval, the new price is shown against the old one for confirmation. Nothing is charged on a price you did not see.

Cancelling marks it declined and Sparky asks what you would like to change instead. The decline is carried into the next message, so it does not silently propose the same thing again.

A proposal left undecided expires after seven days and the card dims.

Content in the conversation

Richer output opens a panel beside the transcript rather than being flattened into text: a design preview, a sortable data table, an HTML preview, a code block. The divider between the two is draggable, and on a narrow screen the panel takes the whole width instead.

A carousel being written shows Slides so far, each slide appearing as soon as it is final, with its headline, body and image prompt.

A finished run offers Save to Library and a link to its full run page, where the images are approved and the result can be scheduled or published.

Teaching it

Three things in the conversation feed back into what Sparky knows.

Action	What it does
Pin a message	Hovering a message reveals a pin. A pinned fact is sent with future replies, and only the most recent few are, which the brand memory panel says plainly.
Rate a reply	The thumbs under an assistant message. These accumulate into learned preferences: tones preferred and avoided, formats, words to leave out.
Correct it in learning	Anything learned can be removed on the learning page. Fixing a wrong fact there is faster and more permanent than arguing with it in conversation.

Conversations

The clock icon opens recent chats, where a conversation can be reopened, renamed or deleted. New chat starts a fresh one.

A long conversation is compacted in the background so it keeps its thread without carrying every message. While that happens the conversation shows as compacting.

While brand discovery is running, chat is paused for that brand and the button says so. It comes back when discovery finishes.

Worth knowing

Drag a file anywhere onto the chat, or paste it. Images, video, audio, PDFs, documents and spreadsheets are all accepted.

Pasting an Instagram, TikTok or LinkedIn profile link offers to analyse that profile rather than treating it as plain text.

While a reply is arriving the send button becomes a stop button. Stopping puts your message back in the composer so you can edit and resend.