.txt

Overview

.txt is an AI-infrastructure company focused on structured generation, also known as constrained decoding, for large language models. Its software makes model outputs conform to schemas, grammars, regular expressions, and function-call formats during generation rather than relying on post-processing, retries, or defensive parsing.

The company is known for maintaining Outlines and for offering hosted and self-hosted infrastructure aimed at production AI applications, including data extraction, classification, agents, and synthetic-data generation. Its products target developers, inference providers, and teams operating model servers such as vLLM, SGLang, and TensorRT-LLM.

History

Founded in 2023 by Rémi Louf, Dan Gerlanc, and Brandon Willard. The founders previously collaborated at Normal Computing on methods combining statistical modeling and compiler technology to make language-model outputs more reliable. The team created Outlines, an open-source library for structured generation, before expanding into commercial infrastructure for constrained decoding, APIs, and inference-serving integrations. The company remains privately held and independent; no acquisition was identified.

Product Lines

Manufacturing

As a software infrastructure company, .txt does not manufacture physical products. Its products are developed as proprietary and open-source software, delivered through APIs, libraries, and integrations with self-hosted inference servers.

Notable Products

• Outlines - Open-source Python library for structured generation and constrained decoding across language models and inference stacks.
• api.dottxt.ai - Hosted, OpenAI-compatible API for generating outputs that conform to JSON Schema, regular expressions, and other constraints.
• dotjson - Structured-generation engine for enforcing JSON Schema constraints during model decoding.
• dotgrammar - Library for generating text under arbitrary context-free grammar constraints.
• dotlambda - Library for reliable function calling and structured tool outputs in agent workflows.

Reputation

.txt is regarded in developer and AI-infrastructure circles as an early specialist in structured generation, with Outlines providing an open-source foundation and broad compatibility across model providers and serving stacks. Its approach is valued for reducing invalid outputs and integration work in production systems. The company operates in a competitive area as major model providers increasingly offer native structured-output and tool-calling features. Public information about commercial pricing, customer scale, and independent performance comparisons remains limited.