Kiln

Kiln is a free desktop app and open-source Python library for teams building production AI. It unifies the entire AI development lifecycle — prompt engineering, evals, RAG, agents, fine-tuning, synthetic data generation, and dataset management — into one local-first workspace. Built on a 'measure, then improve' philosophy, Kiln uses evals as the central metric to drive optimization, including an auto-optimizer that tunes prompts to eval scores. Designed for engineers, data scientists, and subject matter experts (PMs, QA), Kiln lets engineers deploy via its MIT-licensed Python library, data scientists run experiments, and non-coders contribute ratings, feedback, and golden datasets. A built-in AI Assistant converses with users to help set up experiments, create evals, and optimize configurations. Key features include prompt management with versioning, an evals platform with LLM-as-Judge and automated eval builder, synthetic data generation (filter, label, generate), fine-tuning support (distill into smaller models), RAG (index, chunk, retrieve, rerank, semantic chunking), agents with sub-agents, skills, tools, and MCP support, and an Auto-Optimizer. Git-backed versioning for datasets and evals enables async collaboration. Kiln works with 190+ models (cloud and local). The desktop app is source-available, the library is MIT licensed, and the company charges for cloud-based Kiln Pro and enterprise tiers, while the core app and library remain free for individuals. Compared to alternatives like LangSmith or Weights & Biases, Kiln offers a more integrated local-first approach with evals-driven optimization.