How to Use ChatUp: Models, Assistants, and Tools

ChatUp combines general chat, model choice, specialist assistants, and tools for images, voice, files, video, research, animation, and goals.

This guide explains where each part fits. Exact model names and availability can change, so use the current in-product catalog for the live lineup.

Start with prompts, because everything downstream depends on them

Beginner: say what you want, in context

The single biggest jump in output quality comes from adding three things to any request: who it is for, what it is for, and what constraints apply.

Instead of “write an email to my landlord,” try:

Intermediate: assign a role and a format

Advanced: make it interrogate you first

One useful technique is to let the assistant gather missing context before it answers:

Choosing a model

The current catalog can include models from OpenAI, Anthropic, Google, xAI, and DeepSeek. The selector shows the live name and availability; attachment controls reflect whether an available model can accept the material you added.

When an answer feels off, try the same brief with another available model. Treat the result as another perspective, not verification: check factual claims against reliable sources even when both answers agree.

The assistants

Specialist assistants provide a named role, a short brief, and suggested prompts for recurring jobs. They are starting points rather than substitutes for the context only you can provide.

For work: Business Planner, Marketing Expert, Interview Coach, Research Assistant, Financial Analyst, Lawyer.

For learning: Study Helper, Language Teacher, Research Assistant.

For life: Chef, Travel Guide, Fitness Coach, Wellness Advisor, Routine Planner, Relationship Coach, News Reporter.

For making things: Creative Writer, Logo Designer, Tattoo Artist.

And: World Cup Expert, for when that is what the week is about.

Use an assistant when the job repeats. Type into a general chat when it does not.

Note the boundaries on two of them: the Lawyer and Financial Analyst assistants are for understanding terms, comparing scenarios, and preparing better questions — not for advice you act on without a qualified professional.

The tools

Image Gen

Generate images from a prompt. The selected image model may also accept reference photos when the picker says it does. Specificity helps: name the subject, style, lighting, palette, composition, format, and anything that must remain unchanged.

Animate Photo

Turns a still photograph into a short video when the animation service and selected effect are available. Use only photos you have the right and permission to animate.

Voice Chat

Hands-free conversation. It can be useful while cooking or walking, and for language practice where speaking is the skill and typing is not. Do not use it when interacting with a screen would be unsafe.

YouTube Summary

When an available model publishes YouTube capability, submit a supported YouTube link for a summary and follow-up questions. Verify important details against the video.

File Assistant

Work with your own documents: contracts, reports, research papers, statements. Ask targeted questions rather than “summarize this,” and verify anything numerical against the source. Tables and scanned pages are where document extraction is weakest.

AI Research

When the connected catalog publishes an available research tier, use it to retrieve current sources rather than relying only on model recall. Open the sources before repeating a claim.

AI Goals

Create an account-synced goal with a cadence and time, review the generated plan, and follow its scheduled message history. Enrolled browsers can receive Web Push when goal reminders are enabled.

Working with your own material

The pattern that changes the most for people: stop asking about the world and start asking about your material. A contract you were sent, a report you have to present, your own notes, a photo of a machine you cannot operate. That is where the answer is one you could not have got from a search engine.

The habits worth building

1. Ask it to ask you questions. Consistently the highest-return instruction.
2. Switch models when stuck rather than rephrasing endlessly.
3. Verify anything with a number, a date, a price, or a legal consequence. Fluency is not accuracy.
4. Ask for the argument against. Assistants agree too readily; requesting the strongest objection is how you get past that.
5. Rewrite the final version yourself for anything that goes out under your name.

Frequently asked questions

Which model should I use?

Start with an available general chat model. If the task matters, compare another available model using the same brief and evaluate both against your source material and constraints.

Do I need to learn prompt engineering?

No. Add context, state constraints, and ask it to ask you questions. That covers most of the value; the rest is diminishing returns.

Can I reopen previous conversations?

Yes. Signed-in conversation history syncs with the same account across browser and iPhone. A new conversation still needs its own relevant context.

Can I use it on my phone and my laptop?

Yes, and conversations sync, so something started on one continues on the other.

Try the tool that matches the task

Use File Assistant for a document you need to inspect, AI Research for a current question that needs sources, or AI Goals for a plan you want to revisit. Start with low-risk material and verify anything consequential.