Skip to main content
End-to-end recipes. Across every one: ContentHero supplies the context and the execution, the host LLM does any drafting (grounded, never generic), and the user approves before anything ships.

​
Ground the workspace on a brand (run this first)

The setup every other recipe assumes.
1

Resolve the brand

list_brand_kits to find the active kit (one, the default, or ask).
2

Read the voice

get_brand_kit and offer the voice back in a line.
3

Cache it

Write active_brand_kit_id (and any default avatar and connected accounts) to .contenthero/context.md so later steps do not re-ask.

​
On-brand caption for an existing reel

The everyday request: a clip, and a caption in the user’s voice.
1

Ground

list_outliers (scoped with brandKitId), then get_inspiration_content on the top few for hooks and structure. get_brand_kit for voice. get_brand_account_performance for the user’s own best posts. search_brand_knowledge for the brand’s stance.
2

Draft (host LLM)

Write the hook and caption emulating the user’s proven patterns, in their voice. Present for approval.
3

Attach

create_post (approved caption as the description), add_post_asset (the reel), add_post_destination.
4

Ship

schedule_post or publish_post, after the user confirms.

​
Outlier to on-brand clip, scheduled (the full pipeline)

Research to published, end to end.
1

Ground

list_outliers, then get_inspiration_content (patterns), get_brand_kit (voice and visual), search_brand_knowledge (stance), get_brand_account_performance (own top posts).
2

Draft (host LLM)

Concept, hook, and caption grounded in the above. Approve.
3

Produce

generate_image for the key frame, then chain its output id into generate_video as the start frame. Ground the visual in the brand’s visual style.
4

Assemble

create_post, add_post_asset (the video), add_post_destination (platform, format, connected account).
5

Ship

schedule_post, after confirming destinations and timing.

​
Talking-head post

A presenter clip from a script the user supplies.
1

Voice the script

generate_audio (a TTS model, the user’s script, a voiceId). Capture the audio output id.
2

Lip-sync

generate_lip_sync (a portrait image plus the audio output id) for the talking-head video.
3

Assemble and ship

create_post, add_post_asset (the video), add_post_destination, then schedule_post.
The script is the user’s. If they have not supplied one, that is a grounded pipeline Draft, not an invention.

​
Capture a lesson into the knowledge base

How the brand gets smarter over time.
1

Notice the moment

After a decision, a result, or a shared resource, offer to remember it.
2

Add it

add_brand_knowledge (sourceType: text for a lesson or decision; url or youtube for a resource; file or fileUrl for a document or media). Confirm first, give it a clear title.
3

Confirm

Tell the user it is now searchable. Future Ground steps will surface it via search_brand_knowledge.