Brands

A brand is Genfeed’s content context. Brand records can hold identity, guidelines, voice/tone, assets, and provider-facing generation context.

Supported generation flows can apply configured brand context. The exact fields and automatic behavior depend on the current UI and workflow node; these docs do not claim that every setting is applied to every generation.

Cloud organizations can contain multiple brands. Community also supports multiple brands inside its single organization.

Use the product’s brand settings as the source of truth for the fields available in the deployed release.

Brand voice from your own posts

Brand settings → Brand voice → Generate drafts the brand’s writing voice: tone, style, audience, messaging pillars, topics, writing rules, real example posts, prompt seeds, and conversation starters. The agent’s draft_brand_voice_profile tool produces the same draft in chat, where you review it on a card and approve it before anything is saved.

How the brand writes comes from its own posts. How it is positioned comes from the website and the stored brand details. Where they disagree on tone or style, the real posts win. The draft reads, for the current brand only:

• Imported account history. Turn on Import existing posts in Brand settings → Integrations → Imports before connecting an account, or run the import later from the same page. Only the brand’s own connected accounts are read. Reposts, quote posts, and posts written by other accounts are skipped. Replies are kept and measured separately from original posts.
• Posts Genfeed published for the brand. They count slightly less than posts the account wrote itself, because Genfeed drafted them.
• Posts you paste. In agent chat, paste 10–20 of your own posts or replies and ask the agent to draft your brand voice. Pasted text is used verbatim; the agent must not write or paraphrase samples for you.

From these posts the draft measures habits such as typical length, casing, final punctuation, emoji and hashtag use, em dashes, line breaks, and how replies open or disagree. Clear signals become concrete writing rules (“Keep replies short”, “Do not use hashtags”). The example posts saved with the voice are always copied word for word from your real posts, never rewritten by the model.

With fewer than 10 of your own posts, the draft still saves, but the page and the chat card warn that it will not sound like you yet and explain how to add more posts.

Approving a draft in chat only writes the fields the draft actually filled. A list you edited on the Brand voice page is not cleared by a draft that left it empty.

What the agent knows

Brand settings → Agent context shows, read-only, everything a chat turn in this brand is given: identity, guidelines, voice, strategy and topics, custom instructions, prompt seeds and starters, performance insights, proven patterns, retrieved Knowledge and saved context, recent posts, saved memories, active skills, the chat model and its cost per round, the context budget, and the raw system prompt. Each layer shows whether it is in the prompt, stored but not in the prompt, or empty, and links to the page where you edit it.

Type a message into Preview a message to see what a turn about that topic would retrieve. Knowledge, related posts, and memories are ranked against it. Nothing is saved and no credits are charged.

The page also lists your personal memories, which only you can see and can archive, and the brand’s shared memories.

The same snapshot is available to agents as the get_brand_context tool, in agent chat and over MCP, and through GET /v1/brands/:brandId/agent-context (optional query parameter for the preview message).

Knowledge in agent chat

In Brand settings → Knowledge, sources with the Brand Truth purpose that belong to this brand, or are shared across the organization, are added to agent chat as Brand Knowledge when they are relevant to the message. The agent treats them as verified facts. Inspiration and Research sources are never added automatically; the agent reads them only when it calls the search_knowledge tool. Another brand’s Knowledge and saved context never reach this brand’s prompt.

Brand context has a fixed character budget. When it overflows, lower-priority sections are trimmed first: retrieved saved context, then recent posts, performance history, Brand Knowledge, and general context. Custom instructions, guidelines, and voice are trimmed last, in that order. The Agent context page shows what was trimmed.

Edits to the brand’s profile, guidelines, voice, or interview answers reach the next chat turn immediately; the agent does not keep serving a cached copy of the previous brand context.

Watermarked client previews

In Brand settings → Brand Kit, configure watermark text, choose an uploaded brand logo or reference image, and set opacity and corner position. Text and a logo can be combined. The settings are available in Cloud and self-hosted installs.

On an image or video, open the arrow beside Download and choose Download with brand watermark. The export creates a separate PNG image or MP4 video and keeps the original unchanged. The main Download action continues to download the original. You can send the preview to a client and deliver the original after collecting payment. Payment collection and automatic original unlocking are not included.

Watermarked exports require media already saved in storage and a configured brand watermark. Logos must be still PNG, JPEG, or WebP images from that brand. Animated image exports are not supported. Video rendering may take several minutes.

For operators, apply the brand watermark database migration and configure the same GENFEEDAI_API_KEY internal service secret on the API and files services. The renderer rejects requests when its secret is absent or incorrect. Both local and S3 storage providers are supported. Self-hosted operators must configure GENFEEDAI_CDN_URL with the browser-reachable files-service origin that serves /local/; an internal Docker hostname is not a browser URL. The API allows up to ten minutes for an export; configure your reverse proxy to allow the same request duration. Exports are limited to five per user per minute and two active renders per files-service instance. Identical source content and watermark settings reuse the existing derivative.

This optional preview feature does not restrict original access by subscription. A SaaS paywall requires protected original storage and access enforcement across all media delivery paths; a watermarked preview alone is not access control.