Copenhagen-based Palette is building an operating system for companies beginning to work with several AI agents at once. Its premise comes from a practical weakness in today’s AI stack: models can improve quickly, yet they still arrive without the accumulated understanding of a company’s projects, decisions, responsibilities, files, and working habits. Palette wants that organizational memory to remain with the business, even when the model doing the work changes.
The company has raised €3 million in pre-seed funding, led by Ugly Duckling Ventures, with participation from Emblem, Acadian Ventures, and angel investors. Palette says the capital will support a wider rollout of Palette Desktop, continued development of Palette OS, team expansion, and growth from its Copenhagen base as it moves from early adopters toward broader commercial use.
Founded in 2025 by Lars Ettrup, Steffen Sommer, Brian Kyed, and Christian Lomholt, Palette brings together experience from companies including Linkfire, Monstarlab, Podimo, and Planday. The product grew out of a problem the founders encountered while putting AI agents to work across engineering, operations, and commercial functions.
Each agent could be capable, yet useful context kept disappearing between sessions. Teams repeatedly had to explain what a project was, which decisions had already been made, who owned what, and how the organization preferred to work. The intelligence was improving faster than the memory around it. That gap became the foundation for Palette.
Its first product, Palette Desktop, is already available on Mac, with Windows planned. It allows teams to run agents including Claude Code, Codex, Gemini, Mistral, and Palette against the files and folders they already use, while keeping changes reviewable before they are saved. The larger idea is flexibility: teams can choose different agents for different jobs without rebuilding their working environment around a single model provider.
Palette OS expands that concept from individual workspaces into shared organizational infrastructure. Palette is developing a living context library, reusable AI skills, connections to company tools, and handoffs between agent sessions, with parts of the system already being tested with design partners. Compatible AI tools can access that shared context through MCP, allowing knowledge created in one workflow to become useful elsewhere.
That leaves Palette with an ambitious infrastructure bet. As companies add more agents, the difficult problem may gradually move away from choosing the smartest model and toward preserving what the organization already knows. If AI-native teams become increasingly multi-model, Palette wants the durable layer to be the context those models inherit, rather than the model companies happen to use today.
