The Simular AI agent wants to operate your Mac or Windows PC for you

simulationa startup that builds AI agents for Mac OS and Windows, has raised a $21.5 million Series A led by Felicis, with NVentures (Nvidia’s venture arm), existing seed investor South Park Commons, and others joining in.

Simular is an interesting startup because, unlike others, it doesn’t try to control the browser, but rather the computer itself. (Agent AI refers to systems that can complete complex tasks autonomously with minimal human intervention.) “We can literally move the mouse across the screen and do the clicking. So it’s more capable of doing, and replicating, any human activities in the digital world,” co-founder CEO Ang Lee told TechCrunch, giving the example of copying and pasting data into a spreadsheet.

On Monday, it announced the launch of its version 1.0 for Mac OS. But it is also working with Microsoft to develop an agent for Windows. The startup is one of five agent companies accepted into the Windows 365 for Agents program Microsoft announced it in mid-November. (The others are Manus AI, Fellowu, Genspark, and TinyFish.) As for the timeline for a Windows release, Li was vague except to say that it promises to be as or more popular than the Mac version.

Another reason to watch Simular is the goodwill of the founders: Li is a lifelong learning scientist who previously worked at Google’s DeepMind, where he met its co-founder, reinforcement learning specialist Jiachen Yang. Although their team published its fair share of research papers, the work was not strictly academic, Lee said. It was intended to improve Google products, including Waymo.

The background of these AI products is useful because before the efficient future of Silicon Valley’s dreams can be realized, there are a host of technical problems to solve. One of the biggest ones is LLMs hallucination Some percentage of the time.

Agent tasks can require the completion of thousands to millions of discrete steps. Not only can a hallucination at any single step can undo all of an agent’s action, but hallucinations become statistically more likely as the number of steps increases.

One way to solve this problem is to make the “non-deterministic” LLM “deterministic”, meaning that instead of allowing the LLM to be endlessly creative, its responses or actions are written the same way every time. But this risks limiting the agent’s entire creative problem-solving side.

Similar to marry Monday. Its agent will freely repeat the task, with the human user correcting the middle path, until the agent achieves success. A human then secures the workflow for that task, making it deterministic and repeatable.

“Our solution is to allow customers to continue exploring a successful path,” Lee explains. “Once a successful path is found, it becomes an inevitable symbol.”

The reason the startup can do this is because its work — which Lee admits is still early — is not just an LLM wrapper that sends and retrieves data to the model.

“We have a new technology that no other agent company uses. We call it ‘neuro-symbolic computer-using agents.’ It’s not completely based on LLM,” he said. “Our approach to solving hallucinations is to let LLM write code that becomes deterministic. So, if you have a successful workflow, the next time we run the same workflow, it will be successful too.”

Another benefit is that this imperative code that performs a repeatable task is in the hands of the end user, not the LLM. “Once they have the code, they can trust it, because they can examine it, audit it, and see what’s going on,” Lee says.

Time will tell if this method is the magic that will bring agents into the hands of every worker. Lee says his early pilot clients include a car dealership that automates VIN number lookups, and HOAs that extract contract information from PDF files. And the company Open source project (Only available for Mac OS at the moment) It has led to automations ranging from content creation to sales and marketing.

Simular previously raised a seed round of $5 million, bringing the total amount raised to approximately $27 million. Other investors in the company include Basis Set Ventures, Flying Fish Partners, Samsung NEXT, Xoogler Ventures, and podcaster and investor Lenny Rachitsky, the company says.

See the latest revelations on everything from agent AI and cloud infrastructure to security and more