Some founders change their location on LinkedIn to larger cities to improve their reach, but not Rémi Louf. Not only does this French founder live in a village a 70-minute drive away from Paris, but he rents office space in the local castle. This, too, can attract attention — and convey the image of an entrepreneur who doesn’t follow the flock.
Louf’s office setup is the pragmatic decision of a fully remote CEO: he lives with his wife and kids in one of France’s prettiest towns, whose picturesque castle happens to rent out coworking space. It doesn’t have A/C, but it has a fiber connection, and is “cheaper than a garage in Palo Alto or an office in Paris.” But the bargain has other benefits.
“When I was in SF a couple weeks ago, I got recognized by X users as ‘the guy from the castle’,” he told The Infinite Loop with a laugh. Louf has been humble-bragging about his ‘Frenchmaxxing’ lifestyle on X, and in a July first-person account for Business Insider titled “My startup has a literal moat, thanks to the French castle I run it from.”
Credits: Screenshot Louf’s X account.
The startup in question is Dottxt, which forces AI models to follow a specific structure in their answers. But the moat may be Louf’s willingness to sail alone. It takes a contrarian mindset to be early to a problem, but also to keep on going after it looks solved.
From upstream to mainstream
The likes of Anthropic and OpenAI now also provide tools to control LLM output, but there were none three years ago when the team behind Dottxt launched open-source library Outlines to make it easier to adopt AI in production. At a time when developers still had to resort to prompt hacking, they used their background in statistics to make AI results more readily usable in workflows — for instance, by turning customer emails into standardized support tickets.
“For a year and a half the field was skeptical. Then everyone copied the surface. Structured outputs became table stakes,” the team wrote. While this made their open-source project very popular, the rising tide could have drowned the startup they had founded in 2023; being first doesn’t always pay off.
Louf acknowledged it wasn’t smooth sailing, and being challenged on its initial use case — data extraction — rocked its boat. “Last year was quite rough,” he said. That changed with the rise of agentic AI, which opened up a double use case for Dottxt: give agents a structured format to communicate with each other and to take action in the real world through function calling, also known as tool calling.
According to Louf, tool calling is only solved on the surface: it is seamless with closed-source models, but not with open models, hindering their adoption despite growing interest in the alternative they represent. To address this, Dottxt has built a new product called dotlambda, and says it also improves token efficiency, a rising concern as AI moves from training to inference.
Surfing the inference wave
The concomitant rise of agents, open models and AI moving into production have turned the tide for Dottxt. “We are seeing a lot of traction around our tool calling product,” Louf said. The startup has bounced back since February, with strong interest from inference platforms operating in the cloud and on device. “We were a bit early, and the market is catching up, which is nice.”
Dottxt may have been early, but it was also lucky: its approach turned out to be better suited for inference, Louf said. Compared to other methods that affect performance, and therefore costs, he thinks Dottxt is better placed to surf this new wave. But Louf being French, his optimism is tempered. “I don’t think anyone has a moat,” he said.
In his view, this also applies to Dottxt; a player with infinite amounts of money could copy its technology. But for Louf, the key is time: he estimates that the startup is already “one year ahead,” and says its team is both talented and moving fast. By the time others can catch up, he hopes it will already be further ahead.
Looking at it differently, this combination of talent and velocity may qualify as a moat, a concept that investors have started to redefine more flexibly for the AI era. Dottxt’s funding suggests it met their criteria: it raised $11.9 million to date, via a $3.2 million pre-seed round led by deep tech VC firm Elaia in 2023 and an $8.7 million seed led by EQT Ventures in 2024.
What was once his intuition is also now backed by facts: “Everyone sees the same data as we do; the use of open-source models is really on the rise.” If this says anything about his ability to spot trends earlier than others, don’t be surprised if it becomes a thing to live the castle life — or at the very least, to work wherever you please.
Top photo: Co-founder and CEO of .txt Rémi Louf. Credits: Louf’s account on X
