Palette raises EUR 3M to make company context portable across AI models

Why it matters

Palette's founders are betting that company context, permissions and workflows will become a durable control point in enterprise AI, even as models and agent interfaces change rapidly.

Palette co-founders Lars Ettrup, Brian Kyed, Steffen Sommer and Christian Lomholt announced a EUR 3 million pre-seed round on August 11 to build an operating layer for companies using multiple AI agents. Ugly Duckling Ventures led the financing, with Emblem, Acadian Ventures and angel investors participating, according to Tech.eu.

Ettrup previously co-founded Linkfire and led the music-marketing software maker through its 2021 Nasdaq First North listing before stepping down as CEO in January 2024. Kyed co-founded workplace screen-sharing provider Airtame and served as its co-founder and co-CEO. Sommer was CTO at Monstarlab, while Lomholt was VP of Product and Design at Podimo, according to Palette's founder materials.

That operating history maps directly to Palette's pitch. The four founders have spent years building and selling workplace software. Their new bet is that the valuable layer in enterprise AI will sit above individual models, preserving the files, decisions, workflows and internal knowledge that organizations cannot afford to rebuild whenever a different model takes the lead.

A founder group betting against model lock-in

The founders designed Palette to separate organizational context from the underlying model. Kyed summarized the thesis as a need to "make the model swappable and make your company context durable" in the Tech.eu report.

That distinction matters because model choice is becoming a procurement variable. A team may prefer one provider for speed, another for complex reasoning and a local model for sensitive data. Palette wants the organization's working memory to remain usable across those choices.

Palette's current customer-facing product is Palette Desktop, a macOS application launched on May 19. Users point the application at local or shared folders and run supported agents on the files inside them. Each session works from a sandboxed copy, allowing users to review changes before saving them back. Parallel chats, change history and restoration are designed to give non-engineering users some of the controls developers already expect from agentic coding tools.

Palette Desktop currently supports Claude Code and Codex, according to Palette's product site. Teams can choose cloud-based models or run a model locally. The introductory Team plan costs EUR 149 per month for unlimited users and credits, while third-party agent subscriptions remain the customer's responsibility.

The flat organizational price encourages companies to spread usage beyond a handful of technical employees. It also puts pressure on Palette to demonstrate value across departments, since the product is pitched at founders, product managers, sales teams, operations staff, designers and content workers.

Desktop first, operating system later

Palette Desktop gives the founders a product they can charge for while they develop the broader Palette OS. That larger system is intended to combine a context library, reusable skills, connectors to workplace applications and handoffs between agent sessions.

Palette's context documentation describes a living organizational map assembled from tools such as Slack, Linear, Notion, calendars and email. Employees and agents can access that context through Palette or the Model Context Protocol, rather than copying background into every prompt.

Palette says elements of the operating system are live with selected design partners. The commercial proof point today is narrower: Palette Desktop is available to paying customers, while the wider operating system remains in guided deployment. The EUR 149 monthly price establishes a route to revenue, though it does not establish the scale or depth of customer adoption.

The architecture also creates difficult governance work. An organizational context layer has to decide which employees and agents can see a sales conversation, product decision or personnel document. Palette's value will depend on the quality, freshness and permissioning of that context, rather than the number of connectors listed on a product page.

The financing buys distribution and time

Palette plans to use the financing for product development, hiring and sales.

The investor group reflects that agenda. Ugly Duckling Ventures backs Nordic B2B software at pre-seed and seed, while Acadian Ventures focuses on software shaping how organizations work. Palette needs operational support because the sales motion reaches beyond AI enthusiasts. Deploying shared context across a company requires security reviews, permission design, employee onboarding and measurable use cases.

Palette is entering the market as model providers expand into the workflow layer themselves. OpenAI's Codex application operates on local folders and includes reusable skills. Anthropic introduced MCP as an open standard for connecting assistants to workplace data, reducing the cost of building basic integrations.

Those developments sharpen the founders' wager. Palette has to own the organizational layer that model vendors treat as an integration surface: shared memory, cross-team workflows, permissions and continuity between agents. The pre-seed financing gives Ettrup, Kyed, Sommer and Lomholt time to prove that companies will pay a separate vendor to hold that layer as the models underneath it keep changing.