Liquid Harness is an autonomous agent developed by Liquid AI for adapting Liquid Foundation Models (LFMs) to specific tasks using plain-English specifications. The platform is designed to streamline the process of building, fine-tuning, and deploying AI models, allowing users to achieve a deployable model from a task description in under an hour. It targets users who may not have machine learning experience, emphasizing accessibility and automation throughout the model customization workflow.
md file, which serves as the basis for all subsequent stages. Liquid Harness constructs a data pipeline tailored to the task, generates synthetic data samples, scores and filters them using an LLM judge based on the provided rubric, and manages baseline evaluations. It proceeds through multiple stages—such as data generation, filtering, supervised fine-tuning (SFT), direct preference optimization (DPO), evaluation, and checkpointing—handling each as a distinct, inspectable component. The process is largely hands-off, especially with the --auto flag, which enables the agent to run the full pipeline in one command. The tool either produces a model checkpoint that surpasses the baseline or returns a clear failure reason, ensuring transparency and reliability.
Liquid Harness is delivered as a command-line interface (CLI) tool. Installation and operation are performed via terminal commands, and the platform supports concurrent use of multiple GPUs, with the number depending on the selected plan. 2B-Instruct, and the tool is positioned as the official means of customizing Liquid Foundation Models, which are described as small and capable models that can run anywhere.
The service is currently in private beta and requires an invite for access. Pricing is structured around per-seat plans with included monthly credits that cover data generation, scoring, training, and inference. There are Individual, Professional, and Enterprise tiers, each offering different levels of included credits, GPU concurrency, and additional features such as audit logs and dedicated hardware for enterprise users. Usage beyond included credits is billed separately, and all prices are listed in USD. Liquid Harness is maintained by Liquid AI.
Liquid Harness is an Other AI product. It focuses on automating the process of fine-tuning AI models from plain-English task descriptions. It is built as a B2B product for machine learning engineers. The product ships for the command line.
Liquid AI builds and maintains Liquid Harness, and the product first shipped in 2026. Development happens publicly on GitHub with 16 stars and 58 commits in the last 90 days. Key capabilities include model fine-tuning, plain-English spec input, and data pipeline automation. Liquid Harness is currently in beta.
Latest indexed changes and source events
Other apps tracked under the same category.
