Krea connects its creations to Figma, Slack, and Google Drive

Krea launches a creative agent capable of utilizing multiple media, mobilizing specialized agents, memorizing a visual universe, and connecting to services like Figma, Slack, or Google Drive.

Create the visuals for a campaign, adapt them to several formats, prepare a video, and organize the corresponding files from a single request. Krea wants to hand this sequence of operations to Krea Agents, a new workspace capable of managing a creative project using the models and tools available on its platform.

Launched on September 10, 2026, the service can be accessed from the upper-left corner of Krea’s regular interface or at krea.ai/agent. The page can be viewed without an account, but users must sign in before starting a generation.

Krea is not replacing its image and video creation tools with a simple chat window. The agent serves as a coordination layer. It receives an objective, examines the supplied files, determines which operations are required, and then uses the platform’s features to create and modify the different parts of the project.

The prompt box accepts several types of media in addition to text. A request can include reference images, videos, documents, or assets already saved in the workspace. Krea has not yet published a complete list of supported formats, file-size limits, or the maximum number of files that can be attached to a session.

Users can select the large language model responsible for directing the work and adjust how much effort it should devote to the request. The public interface displays an automatic model-selection mode and a medium effort setting. The full list of available models and effort levels is only visible after signing in.

This language model is not necessarily the one that creates the final media. It interprets the request, organizes the steps, and then calls the appropriate generators. A single session can therefore combine one model responsible for reasoning with several image, video, audio, or transformation models available through Krea.

This distinction prevents the agent from being mistaken for another universal generator. Its value lies mainly in its ability to sequence services that previously had to be opened and configured separately. An initial image can be used to create several variations, which can then become shots in a video or assets in a broader campaign.

For simple tasks, a single agent can execute the request directly. Krea says that more complex projects can trigger a team of specialized agents. The main agent then divides the work, delegates certain parts, and brings the results back together within the session.

The word “team” does not refer to several people or necessarily to assistants with permanent individual identities. It describes a structure in which multiple processes can work on separate subtasks. Krea has not yet detailed their roles, how many can be involved, how they communicate, or what criteria determine whether a request requires this kind of organization.

The company has not provided measurements comparing this approach with the use of a single agent either. Additional agents may explore more directions or work on several assets in parallel, but they may also consume more resources and produce conflicting decisions. The quality of the orchestration will therefore need to be assessed on real projects.

The interface supports this structure with a sidebar dedicated to tools. Krea says these tools were designed to provide visual feedback to the agent. Media viewers allow it to examine generated material, while other panels provide access to files and the features used during the session.

This review stage matters in creative work. An agent cannot simply verify that a file was successfully created. It must also examine its composition, consistency across multiple outputs, text legibility, and fidelity to the references. Krea has not specified which checks are automatic or whether every creation is actually reviewed before being shown to the user.

The public page already offers several project templates. They include a bakery rebrand, a fashion film, packaging concepts, a property walkthrough, a product shoot, automotive advertising visuals, a game scene, branded merchandise, and proposed YouTube thumbnails.

These examples provide a clearer picture of the product’s positioning than the broader idea of a creative assistant. Krea Agents is designed for deliverables involving several media types and multiple production stages rather than an isolated generation. A request can begin with research into an art direction, continue with a series of images, and end with an animation or a coordinated set of campaign materials.

The file system plays a central role in this approach. According to Krea, it manages the context the agent uses to understand the user and adapt its work. References should therefore not need to be attached again to every prompt if they have already been organized in this workspace.

This context can contain logos, references, previous versions, brand information, or assets defining a visual direction. In theory, the agent can use them to maintain consistency across several requests and avoid restarting a project from scratch in every session.

Krea has not yet provided a precise technical definition of this memory. It remains unclear which files are loaded automatically, how long they remain associated with the user, how much material can be considered at once, or how outdated information can be removed without erasing the rest of the project.

Poorly managed memory could also produce the opposite of the intended effect. An old brand guide, a discontinued logo, or a reference added for a single campaign might influence later work. Folder organization and the ability to select the context explicitly therefore become as important as the generation itself.

The agent can import a moodboard from Pinterest or other websites. This feature removes the need to download every image manually and allows a collection to be turned quickly into a starting point. Krea does not specify whether the import copies the files, stores only their links, or creates an intermediate representation to guide its models.

That convenience