Kiln
A free desktop app and open-source Python library for building, evaluating and optimizing AI systems, covering evals, RAG, agents, fine-tuning and synthetic data in one workbench.
Kiln lets a whole team, including product managers, subject experts and QA, rate outputs and add data without writing code through its desktop app, while an MIT-licensed Python library ships the same tasks to production. It runs locally with a user's own API keys, or fully offline with Ollama.
An eval builder auto-generates evals, a judge plus a synthetic eval dataset, and aligns them to a team's preferences; an auto-optimize feature searches for the best prompt, model, tools, skills, subagents and parameters for a task; drag-and-drop documents such as PDF, image, video or audio build a RAG with auto-generated evals; subagents compose multi-agent hierarchies, each in its own context window; synthetic data generation supports evals or fine-tuning; zero-code fine-tuning spans more than 60 models on Fireworks, Together and Vertex with serverless deployment; and the app syncs to Git automatically, including for teammates unfamiliar with Git.
The desktop app installs for macOS, Windows and Linux; it is also usable as an open Python library.