- ChatGPT and Claude have projects. Why switch?
- For client work, yes. A ChatGPT or Claude project remembers one space. Underpaint is built for many clients at once: each with its own brand memory and files, kept apart in the database, with methods for four agency roles and the cost per client. The same models are inside. One subscription instead of several.
- Can I use it at my agency without my boss’s approval?
- Check your agency’s AI policy first. Start with your own project, a public brief or a past pitch. Names you mark are masked before anything reaches a model, and files stay in the EU. When it works for you, invite your team or show it to your agency.
- Who owns the output?
- You do, as far as the law allows. You still check claims, trademarks and facts before anything runs.
- Do you train models on my work?
- Underpaint does not. Models sit behind OpenRouter. We turn off providers that train on customer inputs when the account allows it.
- What happens when the AI budget runs out?
- Paid models step down to a faster one, or stop if you picked the model yourself. Fast replies keep working inside the daily cap. You can top up.
- Can I cancel anytime?
- Yes. Access stays until the end of the period you paid. Read and export stay for 30 days after that.
- Is client data separated?
- Yes. One client's files, chats and brand never enter another client's context. The database enforces it, not just the interface.
- Which models?
- GPT from OpenAI, Claude from Anthropic, Gemini from Google, Mistral from Europe, plus Perplexity for web search and image models such as Nano Banana, GPT Image and Flux. The full list is in the Models section, built from the live catalogue.
- Is there one invoice for a team?
- Team and Agency are billed per seat on one subscription. Agency can export cost per client if you recharge.
- Is this legal advice?
- No. We do not review ads for compliance. That is still your lawyer and your client.