Kiln
A local AI workbench for macOS, Windows and Linux. Evaluate agents, optimize prompts and run fully offline with Ollama or use cloud APIs.
Kiln is a desktop workbench for teams building AI applications on macOS, Windows and Linux. It keeps a task and its dataset together across evaluation, prompt optimization, RAG and fine-tuning, so teams can compare changes against the same examples. Engineers, data scientists, QA staff and subject matter experts can contribute through the app.
Kiln runs on your machine and can work fully offline with Ollama. It also connects to cloud providers through your own API keys, including OpenAI, Anthropic, Gemini and OpenRouter, plus OpenAI-compatible endpoints. Cloud model calls go to the selected provider; choosing a local backend keeps model execution local.
Evaluations combine model-based judging with golden datasets and human ratings. The Eval Builder generates judges and evaluation data, while Auto-Optimize searches for better prompts and model configurations against those scores. An AI assistant can create evaluations, run experiments and propose improvements through conversation.
For document-based applications, Kiln supports RAG over PDFs, images, video and audio, with evaluations generated from your documents. Agents can use MCP tools, reusable skills and subagents. Synthetic data generation and fine-tuning share the same task data used for evaluation.
Git sync keeps datasets in your team's repository. The open-source Python library uses the MIT license and runs the app's tasks and project files in production without a rewrite.
