Pimento’s creative expertise transitions into Vibe

Mistral acquires Pimento to enhance the Vibe experience. The French team brings its expertise in creative interfaces, orchestration, and human control.

An assistant capable of generating text, searching for information, or editing code is not automatically easy to use. As its capabilities expand, the interface has to understand the request, select the right tools, and return a result that people can still modify. Mistral is acquiring that expertise through Pimento.

Florent Facq, co-founder and CEO of the French startup, has confirmed that Pimento is joining Mistral. The team will become part of the Engineering and Products organizations to improve the experience of Vibe, Mistral’s unified agent for conversation, professional work, and coding.

The exact value of the transaction has not been disclosed. Mistral only told AFP that the acquisition was worth several million euros. The company has not specified how much will be paid in cash, whether shares are involved, or which conditions apply to the founders and employees joining the company. An acquisition focused on Vibe

The team’s destination is clearly identified. Pimento is not joining the organization responsible for training Mistral’s foundation models. Its employees will work with the product and engineering groups responsible for Vibe’s user experience.

Formerly known as Le Chat, Vibe is now positioned as a unified agent for work and code. Its conversational interface can search for information, draft content, analyze documents, and use tools. Work and Code modes extend that approach to multi-step assignments, software repositories, and professional environments.

That expansion changes the interface problem. A basic chat can be limited to one instruction followed by one response. An agent needs to explain what it plans to do, request missing information, choose tools, flag sensitive actions, and allow users to take back control.

The acquisition suggests that Mistral now sees this layer as being as important as model performance. A capability can be technically available without becoming practically useful if it remains hidden behind menus, settings, or difficult instructions.

Pimento was built around this exact problem. Its product aimed to help creative teams move faster without taking control away from the people responsible for the brief, creative direction, and final approval. From moodboards to advertising production

Florent Facq and Tomás Yany founded Pimento in October 2022. The two founders met at École Polytechnique before pursuing careers combining product development, culture, and machine learning. Facq helped develop France’s Pass Culture program, while Yany worked at Apple and later at Nabla.

Pimento’s first product focused on the beginning of the creative process. Teams could provide a brief, reference images, and brand materials to explore several directions quickly. The platform generated moodboards and visual proposals that could be selected, corrected, or combined.

The goal was not simply to produce a finished image. Pimento was addressing the more uncertain stage when a team has to translate intentions such as “more minimal,” “more premium,” or “more energetic” into references that everyone can discuss.

The startup raised €3 million from Partech, Cygni Capital, and several individual investors. At the time, it described its purpose as accelerating exploration while leaving curation, direction, and judgment in human hands.

The product later moved toward advertising and performance teams. Its platform connected brand identity, previous ad campaigns, and selected campaign data to create new variations for Meta, TikTok, Google, and Pinterest.

It could generate, edit, resize, brand, and animate creative assets while preserving colors, typography, products, and tone. Pimento says it has worked with more than 180 brands, although that figure has not been independently verified. Pix had already anticipated the work ahead

Several weeks before the acquisition, Pimento introduced Pix, a conversational assistant placed above its existing creative suite. Users could brief it in much the same way they would brief a designer.

Pix was designed to review the request, ask questions when information was missing, and coordinate the platform’s capabilities. It could apply brand materials, generate several options, preserve editable layers, and adapt a creative asset to multiple formats.

Pix was not another isolated generator. Pimento already had a collection of tools. The problem was allowing people to use them without learning every menu, setting, or required prompt.

Facq described the change as moving from a professional suite that users had to learn toward a creative partner they could talk to. The assistant did not replace the underlying functions. It decided how to coordinate them and returned the results in a form people could continue editing.

That experience maps directly onto Vibe’s needs. Mistral’s agent also brings together capabilities that do not need to be exposed separately. Someone can request a presentation, a document review, or an application update without knowing which connectors, files, or specialized agents will be required.

Pimento is therefore bringing less in the way of foundation-model technology and more in orchestration expertise. Its team has worked on converting incomplete briefs into sequences of actions, collecting brand context, and returning outputs that remain controllable. Human control as common ground

Facq also frames the acquisition around a shared belief: AI should accelerate work without removing control from the people doing it. That position matches Vibe’s emphasis on human approval for sensitive actions and the ability to move between simple conversations and more autonomous modes.

In advertising, control takes the form of a brief, brand identity, selection among several proposals, and manual editing. In a professional agent, it also concerns which files are accessed, which messages are sent, which project changes are made, and which tools are allowed.

The challenge is to avoid turning human control into a constant series of confirmation windows. If the agent