Brand Rooms

Brand Rooms are live in Create with Chat. The brand is the room. Your team talks in it. Cuppa runs from it — text or voice. Recurring playbooks stay in it.

Strategy used to live in Slack. Drafts lived in Cuppa. Context died in the paste. Brand Rooms keep coordination and the agent in one thread — with brand DNA, keywords, GSC, planner, knowledge, and a live canvas already loaded.

What you can do now

Threads on a brand are shared by default. You see who said what.
Scratch work stays private with the lock (Just me).
Enter asks Cuppa. @ a teammate posts a note (no model call).
Cmd+Enter, Run, or @Cuppa always runs the agent.
One person runs the agent at a time so you do not collide.
Work lands on the canvas (right side): article drafts, grades, SEO optimizer, planner, Performance Hub, PR, images, video, HTML artifacts (films / loops / pages), browser, client reports. Sign off with Approve / Skip.
The agent can write Brand DNA (voice, competitors, visuals), Link Engine, knowledge, and playbooks — not just generate articles.
A good prompt becomes a scheduled playbook (clock in the input bar). Results append to the same thread. Email / Slack when it finishes if you want.
Talk (live voice) sits in the composer. GPT-Live-1 is the default mouth; Grok Voice needs an xAI key. Your chat model still runs the tools.
Ask Cuppa from any page (Cmd+J, floating mic, or Cmd+K Talk to Cuppa). Live voice uses Brand Rooms / the Ask Cuppa panel, not the editor tool subset.
Connected Slack: freeform @Cuppa … and voice memos hit the same full agent. /cuppa generate etc. stay shortcuts.

Talk vs run

Action	What it does
Enter	Ask Cuppa. Agent turn.
@teammate then Enter	Team note. Coordination only. No tokens burned.
Cmd+Enter or Run	Always an agent turn.
@Cuppa	Same as Run — directs the message at the agent.
Talk	Live voice. Same tools as text. You can approve or skip canvas changes by speaking.
@teammate (notify)	Pings that person (in-app; Slack if connected).

The agent treats [Name, note] as background and [Name, to Cuppa] as the instruction. Only the latest to Cuppa turn is executed.

Voice

Talk is a first-class Brand Room client, not a dictation bolt-on.

Provider	ID	Needs
GPT-Live-1	`gpt-live-1`	OpenAI key (default)
Grok Voice	`grok-voice-latest`	xAI key

Live session holds the mic in the browser (WebRTC). MCP/CLI can mint session credentials; they cannot hold the mic on stdio.
While tools run you see short Working on it labels, then canvas updates.
Same agent catalog as typed Chat: research, drafts, optimize, planner, social, PR, video, HTML artifacts, Brand DNA writes, playbooks, alerts.

From any page: Cmd+J, the floating mic, or command palette Talk to Cuppa.

Programmatic: POST /v1/voice/stt, /v1/voice/tts, /v1/voice/session. MCP transcribe_audio / text_to_speech / start_voice_session. CLI cuppa chat transcribe|speak|voice.

Canvas

The room is not a transcript-only chat. The agent docks work on the canvas:

Open and edit article drafts (full editor, not a paste)
Show grades, cost, and the live SEO optimizer (Apply / sign-off)
Planner calendar, Performance Hub, PR Radar, lists
Images, charts, tables — expand and download
Video generation (including Wan 2.7 / 3.0 when WaveSpeed is connected)
HTML artifacts — animated films, looping graphics, single-file landing pages. 5–10 min background build. Download HTML; films/graphics Export MP4 in the tab. See Chat Artifacts
Browser for public competitor / live URLs (not WP/GSC logins)
Client reports — build slides while you talk

Pending canvas changes show a review strip. Say approve or skip (text or voice).

Visibility

Mode	Who sees it
Brand room (default)	Anyone on this brand can jump in
Just me	Only you. Flip the lock in the input bar.

Only the thread creator can change visibility on an existing chat. History shows a lock on private threads.

Scheduled playbooks

When a prompt is worth repeating, click the clock in the input bar (after at least one message):

Name the job (e.g. Weekly content audit)
Keep or edit the playbook prompt
Pick cadence: daily, weekly, monthly, or manual
Optionally notify the team when it finishes
Save

Cuppa appends the next run to the same thread: a scheduled note, then the agent, then the result. The team can reply in the room after.

Empty-state tiles (Morning Brief, Build Content Cluster, etc.) are starter prompts, not scheduled jobs. The clock is what turns a playbook into a cadence. Next run time shows in the room (jobs fire around 06:00 UTC).

Pause, edit, or run-now from the playbook chip on the thread.

API / MCP / CLI: one automation per chat.

cuppa automations create <siteId> \
  --chat-id <chatId> \
  --name "Weekly content audit" \
  --prompt "Listen for brand mentions, then Slack the team" \
  --model claude-opus-5-5 \
  --schedule weekly

MCP: list_brand_automations, create_brand_automation, update_brand_automation, run_brand_automation, delete_brand_automation. REST: GET/POST /v1/sites/{id}/automations, PATCH/DELETE /v1/automations/{id}, POST /v1/automations/{id}/run.

Try this with a teammate

Open Create with Chat
Type the brief and hit Enter (Cuppa runs)
Or @alex We need a cluster around [topic]. then Enter (note only)
Watch the canvas. Approve the optimizer / draft when it looks right
If it is a weekly job, hit the clock and save the playbook
Or Talk and do the same out loud

Why this is not Slack + ChatGPT

Brand-scoped tools (GSC, clusters, publishing, Link Engine, DNA writes) already in the room
Canvas, not a wall of markdown
Live voice with the same tools
Shared threads without a second Slack
Notes (@teammate) do not cost a model call
Recurring playbooks write back into the same thread
Slack @Cuppa and voice memos hit this agent, not a toy command subset

See Agentic AI for the tool catalog.

Notifications

@mentions — in-app (and Slack if connected)
Agent finished — other participants can get a completion ping
Artifact ready — in-app always; email / Slack reuse chat.agent_completed. Failures stay in-app
Playbook finished — email / Slack if you enabled automation.completed

Prefs: Team Settings → Notifications, or GET/PATCH /v1/notifications/preferences. MCP get_notification_prefs / update_notification_prefs / send_notification. Events: chat.mentioned, chat.agent_completed, automation.completed.

FAQ

Can two people run the agent at once?

No. One active agent run per room. Presence shows who is running.

Do notes use my API keys?

No. @teammate notes do not. Enter / Run / Talk / @Cuppa starts a model turn.

Who can see a Brand Room?

Anyone with access to that brand, unless the creator locked it to Just me.

Is this on Solo?

Shared rooms need teammates (Studio+ seats). Solo still gets Chat, Talk, canvas, and can lock threads to Just me. Agentic tool use follows your plan and the Agentic AI table.

Can I run playbooks from a script?

Yes. CLI cuppa automations, MCP brand-automation tools, or REST. Same write path as the clock in the room.

Can I build a film or HTML page from Chat?

Yes. Ask for a film, looping graphic, or single-file page. See Chat Artifacts. Studio+ (agentic). Not on MCP/CLI.

Chat Artifacts — HTML films, loops, pages, in-browser MP4
Agentic AI — Tools, permissions, agentic mode
MCP Server — Voice + automations + Brand DNA writes
CLI — cuppa automations, cuppa chat voice, cuppa brand
Slack Integration — Mentions, voice memos, alerts
Brand Permission — Who can access a brand
Content Planner — Where scheduled content can land

Our take on AI Detectors Chat Artifacts