Kiln
AI Coding ToolsA free desktop app and open‑source Python library for building, evaluating, optimizing, and deploying AI workflows.
Features
- Cross‑platform desktop app for Mac, Windows and Linux with one‑click install
- Eval Builder to auto‑generate evaluation datasets and align results quickly
- Auto‑Optimize that automatically tunes prompts, models, tools and parameters
- Git‑native collaboration letting non‑technical team members rate outputs and add data
- Open Python library (MIT) for deploying agents built in the app
Recent releases
View all 13 releases →- Kiln Chat assistant for AI system optimization
- Automatic Git Sync for parallel team editing
- Library support for streaming responses
Full changelog
Kiln Chat and Automatic Git Sync
What's New
- Kiln Chat: A new chat/assistant panel in the app. Just ask, and Kiln can help you build and optimize any AI system. The assistant can help across all areas of Kiln, including RAG, Tools, Fine-tuning, Evals, Specs, Optimization, and more.
- Automatic Git Sync: Team collaboration has never been easier. Kiln automatically syncs your project over Git so your whole team can make edits in parallel. No command line needed — a simple UI setup flow, then just use the app for instant syncing.
- New Models: Kimi K2.6, GLM 5.1, Gemma 4, Opus 4.7, Qwen 3.6, MiniMax M2.7, Mistral Small 4, and many more.
- And More: Library support for streaming responses, bug fixes, easier signup, and much more.
CI Build Source for this release Mac and Linux: /Kiln-AI/Kiln/actions/runs/24801094423
CI Build Source for this release Windows: /Kiln-AI/Kiln/actions/runs/24801094468
Full Changelog: https://github.com/Kiln-AI/Kiln/compare/v0.26.0...v0.28.0
- Agent Skills standard support for progressive information disclosure
- Thinking level selection for reasoning models
Full changelog
Agent Skills Support
What's New
- Skills: Support for the open Agent Skills standard. Skills allow progressive disclosure of additional information to your agent -- a great middle ground between context loading and RAG.
- Thinking Level for Reasoning Models: You can now select the thinking level for reasoning models that support multiple levels. See the "Advanced" section of the Run screen.
- New Models: GPT 5.4, GPT 5.3 Instant, Qwen 3.5 variants
- And More: Improved error UI, SDK improvements (runs are no longer saved by default), deprecated models hidden in UI, improved fine-tuning for tool use
CI Build Source for this release Mac and Linux: /Kiln-AI/Kiln/actions/runs/23250506105
CI Build Source for this release Windows: /Kiln-AI/Kiln/actions/runs/23250506230
Full Changelog: https://github.com/Kiln-AI/Kiln/compare/v0.25.0...v0.26.0
- Kiln Automatic Prompt Optimizer for automated prompt engineering
- Optimizer Screen for selecting optimization methods
- MCP Run Configurations to integrate external implementations
Full changelog
Automatic Prompt Optimizer & Run Kiln Evals on MCP Servers
What's New
- Kiln Automatic Prompt Optimizer: our new state-of-the-art prompt optimizer. We automatically finds high-performing prompts for your task. It often beats manual prompt engineering by double-digit gains on evals. Instead of human trial-and-error, Kiln will run thousands of automated experiments and iteratively find an optimal prompt for a given model and task.
- Optimizer Screen: Our new menu for finding the best way to optimize your task. Pick from options like prompt optimization, model selection, fine-tuning and more.
- MCP Run Configurations: hook a Kiln task up to any external implementation through MCP. This allows you to use Kiln's evals, specs and compare features with your own custom agent implementations.
- New Models: Gemini 3.1 Pro, Sonnet 4.6, GLM 5, Kimi K2.5, Minimax M2.5, Qwen 3.5, and many more!
- And More: Better app layout/design, featured models, model suggestions, model pricing, and more!
CI Build Source for this release Mac and Linux: /Kiln-AI/Kiln/actions/runs/22231831533
CI Build Source for this release Windows: /Kiln-AI/Kiln/actions/runs/22231831526
Full Changelog: https://github.com/Kiln-AI/Kiln/compare/v0.24.0...v0.25.0
- Tool Use Fine Tuning for model-specific tool training
- Compare Task Configs with interactive charts
- Deploy Kiln Tasks via CLI to custom servers
Full changelog
Kiln Copilot, Specifications, Fine-tuning for Tools and More!
What's New
- Specs & Kiln Copilot: our new agentic system for building better evals. It works interactively with you to refine and improve your evals with synthetic data generation, edge case discovery, judge prompt discovery, and more!
- Tool Use Fine Tuning: Train models to use specific tools, with examples
- Compare Task Configs: new interactive charts to find the best model and prompt for your task using evals
- Deploy Kiln Tasks: we have a new CLI and instructions for deploying tasks you create inside of the Kiln app to your own server or API
- New Models: Gemini 3 Pro & Flash, Opus 4.5, GLM 4.6V and 4.7, GPT 5.2, and many more
- And More: bug fixes, better keyboard navigation, UI improvements, faster tool calling
CI Build Source for this release Mac and Linux: /Kiln-AI/Kiln/actions/runs/21444555657
CI Build Source for this release Windows: /Kiln-AI/Kiln/actions/runs/21444555635
Full Changelog: https://github.com/Kiln-AI/Kiln/compare/v0.23.0...v0.24.0
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.