Kiln AI

About Kiln AI

Kiln AI is a desktop workbench for the unglamorous half of building with LLMs: evaluating, optimising and fine-tuning rather than prompting. It runs locally on macOS, Windows and Linux, and its central idea is that everything a team needs to make an AI system actually work should live in one place — RAG with document indexing, chunking and retrieval; reusable capabilities it calls Skills; tools and MCP composition; sub-agents that hand off and delegate; structured output; and, tying it together, evaluations. The datasets are stored locally and sync through git, which is the design choice that most distinguishes it from hosted evaluation platforms: your eval sets are versioned alongside your code, reviewable in a pull request, and not sitting in someone else's database. It supports more than 190 models. The paid layer, Kiln Pro, runs on the vendor's servers and is aimed at the parts that benefit from a model in the loop — an AI assistant that understands your project, datasets and evals; automatically generated LLM judges and evaluation datasets; and a Kiln Optimizer that tunes prompts for you. At verification the project reported over 10,000 developers and 5,000 GitHub stars. Note that getkiln.ai now redirects here; kiln.tech is the canonical apex.

Key Features

Tags