What Is Palette OS? ㅣAI Agents, Company Data, and Permissions
Palette CTO Hanish Keloth on the Development Philosophy and Future Vision Behind Palette OS
What comes to mind when you think of an operating system? And how far has the concept evolved in the age of AI?
As a company focused on delivering AI experiences, what does Palette mean by Palette OS, and what ideas and principles shaped its development? We explore the answers through an interview with Palette CTO Hanish Keloth.
Palette OS is a company operating system that connects company data, AI agents, permissions, and execution processes in a single workflow.
It goes beyond AI that simply generates answers, aiming instead to be a system that remembers organizational context and carries work through to completion.
In this interview with Palette's CTO, we take a closer look at Palette OS.
Q1. Could you briefly introduce yourself and Palette's development team in India?
My name is Hanish Keloth, and I lead Palette's development team in India as CTO.
Our team is currently evolving from software engineers into product engineers who take responsibility for the product's overall quality as we focus on developing Palette OS. Palette's headquarters in Korea sets the strategic direction, while the India team translates it into architecture and implementation.
Q2. What exactly is Palette OS?
Palette OS is a Company OS that connects information and work scattered across Slack, documents, AI tools, and other systems.
When work tools are disconnected, people have to repeat the same explanations, while the context behind decisions remains with individuals or buried in specific channels. As a result, AI cannot find the information it needs. Palette OS was designed to solve this problem.
For example, the Drive system retains company data, specialized agents carry out tasks, and permission controls manage the scope of information access. Palette OS brings all these components together in a single workflow.
Palette OS is built around four core capabilities:
- Agents: Software that receives a goal, uses the necessary tools and data, and completes a task across multiple steps
- Drive system: A system that stores and manages company data in one place
- Permission control: A capability that ensures users can view and access only the information they are authorized to use
- Natural-language app builder: A tool that enables non-developers to create agents and apps using natural-language prompts and share them with colleagues
Q3. What is the development philosophy behind Palette OS?
Palette OS follows the principle of harness engineering: instead of returning a single answer and stopping there, "the system is given clear boundaries - what it knows, which tools it may use, and where it must stop - and it keeps working inside them until the job is actually finished. Instead of returning a single answer and stopping there, it checks its own output against acceptance criteria and corrects errors in subsequent runs
By repeating tasks, the system learns from errors and corrects them in subsequent runs.
The governance system validates permissions and data, then repeats the task until the output meets the acceptance criteria.
In other words, our goal is to achieve a level of reliability that allows users to deliver the output to clients without reworking it first. A human still gives the final approval - that gate is part of the design, not a limitation
Q4. What sets Palette OS apart from other AI services?
"Palette builds its own Korean-optimized models - language, image, and video - by fine-tuning and further training open-weight bases on our own infrastructure"
However, it does not rely solely on the response quality of any single LLM.
At the core is an architecture that provides long-term memory through individual- and company-level knowledge graphs and Palette's proprietary Drive system. Built on this foundation, Palette OS is designed to keep operations running even when key employees leave. It preserves work records and the context behind decisions, making handovers easier. A roadmap has already been planned through Versions 2 and 3, and individual employees' work records are retained in the OS so successors can continue their work.
Q4-1. What does the Palette Governance OS validate?
Palette Governance OS determines which agent to trigger, validates permissions and data, and then assigns the task. It repeats this cycle until the output meets the acceptance criteria. Only results that pass final approval are delivered to the user, and "user feedback is stored in the system's memory and improves subsequent runs" Ultimately, its greatest differentiator is that the system can proactively drive results without relying on a single LLM.
Image 2 Caption: Example of Palette Governance OS
Q5. What has stood out to you recently while collaborating with Palette's headquarters in Korea?
The day before this interview, I was impressed to see everyone in the lounge connecting their agents to the OS.
Using the Palette SDK, team members in design and marketing built and deployed apps within minutes. Through this process, we realized that while developers may prefer working across multiple tabs, general users want a simple flow: input, submit, and view the result.
That experience stayed with me because it prompted us to think about how we could make the product more accessible to non-developers.
Q6. How do you envision the future of Palette?
Palette's next step is the Palette Second Brain, which connects work context across online and offline environments. It is a system in which wearable devices and meeting bots capture and summarize conversations, then handle follow-up tasks.
The future Palette envisions is not simply a company with more AI tools.
Palette envisions a company where the Palette Governance OS retains context over the long term and uses that context to reduce people's workload.