Liquid AI releases fine-tuning harness for AI agents
Liquid AI introduces Liquid Harness, an autonomous agent that automates data pipelines, sample scoring, filtering, and model fine-tuning from a plain-English description, requiring no ML experience.
AI for AI · A Liquid AI product · Private beta
From zero to a fine-tuned LFM in under an hour.
An AI platform for building better AIs, faster. Liquid Harness is an autonomous agent that takes a plain-English description of your task and ships a deployable model. It writes the data pipeline, scores and filters samples, runs baselines, fine-tunes, and iterates — no ML experience required.
Request beta access How it works
Get started
$ uv pip install lqh # or: pip install lqh
$ lqh --auto ./my-task [stage: rubric] writing scorer from spec [stage: data_gen_draft] 5 samples generated, all valid [stage: filter_validation] 1,427 / 2,000 kept [stage: sft_initial] score 6.8/10 (baseline 4.1) [stage: dpo] iter 3/5, score 7.4/10 [final: success] DPO checkpoint beats baseline by +3.3
Fine-tuning is one step. lqh does the other eight.
One command runs the full pipeline. Each stage is a real component you can inspect, stop at, or hand off.
spec
rubric
data gen
filter
baseline
SFT
DPO
eval
checkpoint
Specify in plain English
The agent interviews you about the task and writes a SPEC.md that drives every downstream stage. No DSL, no boilerplate, no ML jargon.
Synthetic data, scored & filtered
lqh authors a per-task data pipeline, generates samples concurrently, and scores each one with an LLM judge against your rubric. The dataset that hits training is already curated.
Hands-off with --auto
Point lqh at a directory and walk away. It either delivers a checkpoint that beats baseline or returns an explicit failure with the reason — never a hang, never a prompt.
Watch lqh ship a model in one run.
A short walkthrough — spec to deployable checkpoint, end to end.
Built on Liquid Foundation Models
The first-party way to customize an LFM.
Liquid Harness is built and maintained by Liquid AI as the official tool for adapting Liquid Foundation Models — small, capable models that run anywhere — to your specific task. Default base model: LFM2-1.2B-Instruct.
Customize your first model in an afternoon.
Liquid Harness is in private beta. Drop your email and we'll send an install link.