.txt raises $11.9M to make AI speak computer

Large language models are powerful, but their answers don’t always fit into the workflows engineers rely on. Free-form text is fine for people, but brittle for systems that expect well-structured data. Developers end up writing brittle prompts, post-processors, and retries just to get a valid JSON object.

That’s the problem we’re solving at .txt.

Our approach - structured generation - makes models produce outputs that respect a schema, grammar, or regular expression. Instead of patching broken responses after the fact, the model only generates valid tokens in the first place. In other words: LLMs can finally speak computer.

We started by building Outlines, our open source library for structured generation, which has now been downloaded more than 2.5 million times. Outlines proved that there’s a real demand for reliable model outputs, and showed us how much further we could take the idea.

To accelerate that work, we’ve raised $11.9 million in funding.

We are excited to have the following partners on our side for this journey. This seed round was led by EQT, Elaia, Seedcamp, Common Magic, Kima Ventures, Script Captial, Station F with participation of Roxanne Varza, Julien Chaummond, Bob Van Luijt, Erik Bernhardsson, Rick Lamers, Pierre Valade and more.

This milestone allows us to continue adding new talent to our team and continue our mission of making AI speak the language of every application.

The co-founders,

Brandon Willard & Rémi Louf