Voice and memory

Your agent is only as good as what it knows about you. Voice and memory settings are where you teach it what you're working toward and how you sound, once, instead of repeating it in every chat.

Most of this lives in Settings → Agent in your workspace, and you can set your visual brand by asking the agent. The agent keeps your goals, brand, and voice in mind across every chat, so updates you make apply everywhere: drafts, suggestions, scheduled tasks, and social posts.

Goals

Goals tell your agent what to focus on. You can set up to 20.

Good goals are specific enough to act on:

• "Grow my newsletter to 5,000 subscribers"
• "Publish one deep essay a month"
• "Turn my best posts into an X audience"

The agent uses goals to prioritize. If growing your newsletter is a goal, expect more suggestions about newsletter sends and subscriber growth. If a goal changes, update it, and the agent's priorities follow.

Brand voice

Brand voice describes how the agent writes for your site: posts, newsletters, titles, and descriptions. You have up to 10,000 characters, so there's room to be thorough.

Things worth including:

• Tone. Formal or conversational? Playful or restrained?
• Structure. Short punchy paragraphs, or long-form argument?
• Vocabulary. Words you love, and words you never use.
• Punctuation. Habits worth keeping, especially em dashes. Drafts leave them out unless your voice says you use them.
• Point of view. First person? Do you address the reader directly?
• Examples. A few sentences of your writing that capture the sound you want.

The more concrete you are, the closer the drafts land. "Warm but direct, short sentences, no buzzwords" beats "professional".

Social voice

Social voice covers how the agent sounds on social channels like X and LinkedIn. It's separate from brand voice because most writers sound different in a thread than in an essay.

Use it to set the register for distribution: looser or tighter than your site, whether you use emoji, how you open a thread, and what you'd never post.

Visual brand

Your visual brand is how your work looks: its colors, fonts, spacing, corners, shadows, and a few reference images. The agent reads it whenever it makes something visual, so cover images, social cards, and your custom website come out looking like yours instead of generic.

Ask the agent to show it or change it:

• "What's my visual brand?"
• "Set up my brand based on stripe.com"
• "I don't have a website, make me a brand"
• "Set my brand colors to deep navy and warm cream"
• "Use Playfair Display for headings and Inter for body text"
• "Use this font file for my headings" (with the file attached)
• "Use these two images as my brand references"

A few things worth knowing:

• You can point it at a website. Name a site (yours or one you admire) and the agent reads its palette, fonts, and mood, then shows you a card with what it found. You can also attach screenshots, a logo, or a PDF brand guide instead.
• Or point it at a tool you've connected. If your brand already lives in a design tool, an asset library, a repo, or a docs workspace, connect it and ask the agent to build your brand from there. It reads what the source declares (variables, design tokens, a theme file), so colors and fonts come back exact rather than guessed.
• Nothing to point it at? Ask anyway. The agent builds a few directions from what it already knows about you: your voice, your goals, what your publication says it's about, and your logo if you have one. It shows them side by side, each with its colors and real fonts. Pick one and it saves that, then you refine it like any other brand. It'll say it's working from what it knows rather than reading a source, so expect a starting point you'll want to nudge.
• Re-basing your brand on a new site starts fresh. Ask for a different site later and the previous one's values are replaced, not blended. If the new site offers no artwork the agent can use, it clears the old one's reference images rather than leaving them on a brand that's moved on, and tells you so. Your logo is never cleared this way.
• Fonts come from Google Fonts, or you can bring your own. Name any Google Fonts family, or attach a font file (.woff2, .woff, .otf, .ttf) and ask the agent to use it. Only upload a font you own or are licensed to use, since we host the file on your site and serve it to your readers. When you point the agent at a website, it'll tell you which typeface that site uses. If that family is on Google Fonts, it uses the real one. If it isn't, it picks the closest family and names the original, so you can bring the file yourself. A font on someone else's site isn't ours to serve from yours.
• Reference images do the most work. You can store up to three. They pin your palette, treatment, and mood, and they steer generated images more than any other setting, so pick ones you'd be glad to see echoed.
• You have one logo, and it's your publication's. It's the mark readers see on your site, as your favicon, and in your emails, set under Settings → Publication. The brand card shows that same file, so what you see there is what your readers see. If you don't have one yet and you build your brand from a site, we save the mark found there as your logo and tell you. Change it in Settings if you'd rather use your own.
• A site is a starting point, not a copy. The agent reads a source's palette, type, and treatment for your own work, and what it generates won't reproduce another company's name, wordmark, or slogans. Anything it saves from a site is a placeholder you can swap for your own.

Once a brand is set, it applies to the cover options the agent offers for a post, to social cards, to your custom website, and to anything else it generates. Ask it to change any part of the brand and the next thing it makes follows.

Notes

Notes hold extra context that doesn't fit the other fields: your posting cadence, topics to avoid, ongoing projects, or facts about your niche.

Notes are shared memory. You can write them, and the agent also writes its own notes here as it learns things worth keeping. If you spot a note that's wrong or stale, edit or remove it, and the agent moves on with the corrected version.

You can also add to memory from a conversation. If something comes up in chat that the agent should remember permanently, ask it to save a note.

How it all fits together

You stay in control throughout. Suggestions wait for your approval, and you can steer or stop anytime. See Approvals for how review works.