CI
Package
Meta
Apps

Kiln AI

Iteratively Build High Quality AI Products

Key Features

• 🚀 Intuitive Desktop Apps: Apps for Windows and MacOS. Anyone on your team can contribute to your datasets & feedback to improve quality, without code.
• 🤝 Team Collaboration: Designed for collaboration, with git-based version control for your ML dataset.
• 📊 Rapid Evaluation: Compare various models and data-science techniques with no-code tools.
• 🔄 Iterative Improvement: Continuous dataset improvement with built-in evaluation, RLHF and QA tooling.
• 🧑‍💻 Hackers Delight: An open-source Python library and REST API for data scientists and engineers to deeply integrate if needed.
• 🔒 Privacy-First: We can't see your data. Bring your own API keys or run locally with Ollama.
• 🗃️ Structured Data: Build AI tasks which speak JSON. Build training datasets with validated formats, human feedback, issue tracking, and more.
• 💰 Free: Our apps are free to use, and our library is open-source.

Download Kiln Desktop Apps

The Kiln desktop app is a free download, available on MacOS and Windows.

Install Python Library

Our open-source python library allows for deep custom integrations. Read the docs.

pip install kiln-ai

What is this for?

Kiln AI is for building high quality ML agents for products. What makes product development different than standard data science is that it's not static; no perfect dataset exists to learn from. Products are constantly evolving; bugs emerge from users, product goals change, new use-cases are added.

To solve this, Kiln makes it easy for teams to iteratively collaborate on high quality datasets, and quickly evaluate a range of AI foundation models and techniques.

Kiln helps your whole team continuously improve your dataset.

Our apps make it easy to continuously iterate and improve your dataset. When using the app, it automatically captures all the inputs/outputs/model parameters and everything else you need to reproduce an issue, repair the result, and capture high-quality data for multi-shot prompting or fine-tuning. QA and PM can easily identify issues sooner and help generate the dataset needed to fix the issue.

Our apps are extremely user-friendly and designed for non-technical users. This includes one-click to launch, no command line required, no GPU required, and a consumer-grade UI.

Kiln makes it easier to try new models and techniques.

For experimentation, Kiln includes no-code data-science tools to quickly try a variety of approaches and models in a few clicks.

These tools leverage your dataset, and the quality gets better the more you use Kiln. For example, multi-shot prompting finds the highest-rated results to use as examples.

Currently, we support about a dozen models and a variety of prompting solutions (few-shot, multi-shot, chain of thought), with plans for more models and complex techniques (fine-tuning, RAG).

Deep integrations with our library and REST APIs.

If you have a data science team and want to go deeper, our Python library has everything you need. Ingest the Kiln data format into your pipeline. Create custom evaluations. Use Kiln data in notebooks. Build fine-tunes. Extend or replace any part of the pipeline.

We also offer a REST API package for integrating Kiln into your own tools and workflows.

Contributing & Development

See CONTRIBUTING.md for information on how to setup a development environment and contribute to Kiln.

Licenses & Trademarks

• Python Library: MIT License
• Python REST Server/API: MIT License
• Desktop App: Available as a free download under our EULA. The source code is source-available, and under a proprietary license.
• The Kiln names and logos are trademarks of Chesterfield Laboratories Inc.

Copyright 2024 - Chesterfield Laboratories Inc.