AI agents and skills reference
This page lists the AI skills configured under Platform Admin → AI Agents / Skills, the model each is assigned to, and the platform data they draw on. Use it when deciding which model integration to assign to a skill.
Skills
Skills run in sequence when a user launches a campaign with the AI assistant: capture the request, write the brief, generate storyboard and copy, then generate creative for every enabled touch-point. Each skill is assigned an integration from AI Models / Integrations, which holds the provider's API key. When the brief is written by hand, the 2 brief-stage skills are not used. The remaining skills run only when Generate Storyboard & Copy, Generate Campaign Preview or Generate Creative Assets is selected on the campaign.
| Skill | Stage | Produces | Model in the demo tenant |
|---|---|---|---|
| Capture Campaign Details (from User Input) | Brief | Structured campaign details from a typed or spoken request | GPT 5.2 (gpt-5.2) |
| Write Campaign Brief | Brief | Campaign details, promoted product, target audience, objective and touch-points | GPT 5.2 |
| Generate Storyboard & Copy | Storyboard & Copy | Headline, sub headline, body copy, CTA copy and legal copy | GPT 5.2 |
| Campaign Preview Requirements | Storyboard & Copy | Requirements for the campaign preview | GPT 5.2 |
| Campaign Prompt - JSON Formatting | Creative | The structured prompt passed to the image model | GPT 5.2 |
| Generate Campaign Creative | Creative | Campaign images per playlist and orientation | Nano Banana 2 (Gemini) |
| Replicate Background Images (remove copy) | Creative | Background image variants with text removed | Nano Banana 2 (Gemini) |
| Generate HTML tags for a text overlay. | Creative | HTML for the text overlaid on the creative | GPT 5.2 |
| Generate a scene image prompt optimized for video generation. | Video | Scene prompt for a video campaign | Nano Banana 2 (Gemini) |
| Generate image criteria for a video scene frame. | Video | First-frame and last-frame criteria for a scene | GPT 5.2 |
| Generate Preview Video | Video | The preview video for a video campaign | LTX 2.5 (ltx-2-5-pro); Veo 3.1 (Google Gen AI, veo-3.1-generate-preview) can be assigned instead |
Tabs
| Tab | Contents |
|---|---|
| System Prompts | The skills table: Name, Role, Responsibilities, AI Model / Integration |
| Vector DB | The vector store the skills retrieve from |
Inputs the skills draw on
| Input | Where it is set | Used for |
|---|---|---|
| Brand guidelines (PDF) | Company Settings → Brand / Logo / Guidelines — Upload your brand guidelines for the AI agents | Tone, style and compliance requirements in the brief and creative |
| Brand name and primary business category | Company Settings → Brand / Logo / Guidelines | Industry-specific compliance in the brief |
| Target audiences (personas) and campaign objectives | Company Settings → Campaign Settings | Brief, targeting, scheduling and creative optimisation |
| Reference images and style guides | Campaign Storyboard & Copy tab | Creative generation for new products |
| Historic campaign performance | Recorded by the platform | Which messages, offers and creative have performed for the product, objective and audience |
| Real-time context: local offers, inventory, pricing, weather, time of day, foot traffic | Connected data sources | Campaign decisions at playback |
Where the AI runs
Every skill calls its assigned model through your own integrations, with the API keys you registered, and every asset generated and signal captured stays inside your organisation's AWS account.
Related
Last updated 25 September 2026
