Palette Raises €3M to Build an OS for AI-Native Teams

Copenhagen-based Palette has closed a €3M pre-seed round to build what it calls an operating system for AI-native organisations — one that keeps company context durable while letting teams swap AI models freely. The round was led by Ugly Duckling Ventures, with Emblem and Acadian Ventures also participating. The startup, founded in 2025, already has paying customers on its desktop product and design partners testing its broader OS platform.

The Problem: Models Are Temporary, Context Is Not

Every enterprise AI strategy today is quietly hostage to a single vendor. Switch from one frontier model to another and you risk losing the accumulated workflows, internal knowledge, and fine-tuned prompts your team spent months building. That's the organisational debt Palette is betting most companies haven't yet realised they're accumulating.

The Copenhagen startup's stated mission is to build the OS for AI-native organisations — and with a €3 million pre-seed round now closed, it has the runway to make that bet real.

The Round

Palette's pre-seed was led by Ugly Duckling Ventures and joined by Emblem, Acadian Ventures, and a cohort of angel investors. The capital will go toward expanding the sales team and accelerating platform development.

The company was founded by Brian Kyed, Lars Ettrup, Steffen D. Sommer, and Christian Lomholt and incorporated in 2025 — making this an early but deliberate bet on a gap that's only becoming more visible as enterprise AI adoption matures.

What Palette Is Actually Building

The core insight is architectural: separate the AI model layer from a company's internal context, so that switching models never means losing institutional knowledge.

"There will always be a newer model. The responsible way to adopt AI is to make the model swappable and make your company context durable. That's what we're building with Palette." — Brian Kyed, co-founder

Palette's product stack has two components at different stages of readiness:

Palette Desktop — currently available to paying customers, this lets teams work with AI agents like Claude Code and Codex directly on local files and folders. Users can toggle between cloud-hosted models (optimised for speed, cost, or capability) and local models where data sensitivity demands it.
Palette OS — the broader platform under active development with design partners. It includes a company-owned context library, shared AI skills, and a connector gateway — the idea being that a tool is wired up once and then becomes available across all agents in the organisation.

The connector gateway is worth flagging specifically. One of the most friction-heavy parts of deploying multiple AI agents in parallel is the integration overhead: each new agent potentially needs its own connections to Slack, Notion, CRM systems, and so on. Palette's architecture is designed to eliminate that redundancy.

Why This Matters for Operators

AI adoption is no longer confined to engineering teams — and that's precisely where the challenge gets messy. Sales, operations, and product teams are reaching for AI tools built for individual use, not organisational coherence. The result is a fragmented stack where every team member's workflow is slightly different, knowledge doesn't accumulate, and model upgrades require rebuilding from scratch.

Palette's framing — durable context, swappable models — is a direct answer to that organisational entropy. For startup founders and growth operators building AI-forward teams, this positioning is significant: it suggests a future where the switching cost between, say, GPT-5 and Claude 4 is close to zero, because your company's knowledge layer lives entirely outside any one provider's walls.

The Broader Context

This raise lands at a moment when the agent-coordination problem is attracting serious capital across the market. The infrastructure question of how enterprises manage, orchestrate, and retain context across multiple AI agents is becoming the next major battleground — and Palette is staking out an opinionated position on that question before the market fully consolidates around an answer.

For a 2025-founded company to already have paying customers on one product and design partners on a second is a credible signal — even if the OS layer still has a long road ahead.