AI agent in your project

Agent is a project mode next to Generation and Editor. You describe the task in plain words, the Agent picks a model, writes a prompt for it and brings a launch card. The result is saved in the same project.

Each step is visible in the reply: Picked a model, Composed a prompt, Prepared a proposal. The card shows the model, the full prompt, the settings (format, resolution, duration, sound, number of outputs) and a Why this model explanation. The price is shown on the Generate button. Adjust moves the launch to the prompt bar, and the cross rejects the proposal.

A card is valid for 15 minutes. After that, the button changes to Refresh price: the price is recalculated before launch, and nothing is charged on outdated terms.

The launch mode is set for each chat. The default is Confirm: every launch goes through a card with the price. In Auto, generation starts right away. Hybrid launches cheap runs on its own and shows expensive ones as a card. In both automatic modes the Agent launches no more than three generations per reply, and a proposal without a known price always waits for you.

Besides images and video, the Agent launches music and voiceover the same way, through a card with the model and price. If a generation fails, Fix & retry makes the Agent rephrase the prompt and propose a new run. A finished image can be changed with Edit with mask, or you can ask for Another variant right in the chat.

You can attach up to five files to a message: photos, video and audio from your device, project materials, a sketch, or a PDF, DOCX, TXT, MD or CSV document up to 20 MB. The Agent reads documents and uses media as references for the launch.

The Agent receives the project context in every reply: the story, style details and prompt rules from Project settings. Facts and rules are added to them: notes that apply only in the chat and do not affect generation. A project can have several chats. When a chat gets long, the Agent warns which earlier messages will no longer be included in the reply.

For narrow tasks there is the Agent team: eight employees, Card Designer, Product Photographer, Copywriter, Videomaker, Sound Designer, Prompt Engineer, Meaning Maker and Creative Strategist. Each has a separate chat and its own skills, working methods that are connected for the task. A slash command with the employee's role followed by a brief opens that employee's chat and sends the brief right away. You build your own employee or skill with Create employee and Create skill.

The Agent's model is chosen in the message field, from Claude, ChatGPT, Grok and Google models. The price of a reply depends on the model and the actual amount of work, and the model and the amount charged are shown under each reply. Some models reply for free, and their price in the picker is zero. With a paid subscription, Bonus tokens for Agent replies are added every day, and they are spent before your balance.