Introducing Kiln Agent Builder

Kiln Agent Builder

Kiln now supports building agentic systems! You can build agentic systems with subtasks, tools and memory. All in under 5 minutes.

For example, you can built a team of agents to write blog posts with:

• A writer agent that composes content
• A web-researcher agent that searches the web and summarizes results
• A review agent to check the quality of the content
• A coordinator to orchestrate the team

Now, what exactly is an agent? The terms "agent" and "agentic" are often overloaded. The good news is that Kiln supports everything you need to build agents. Learn more about each agentic skill in our docs:

• Tool Use
• Multi-Actor Interaction (aka subtasks)
• Goal Directed, Autonomous Looping & Reasoning
• State & Memory

Context Management Made Easy

Context management is the process of curating the model's context (chat history) to ensure it has the right data, at the right time, in the right level of detail to get the job done.

Dividing your tasks into subtasks with Kiln's new subtask feature makes context management easy. Each subtask can focus within it's own context. This can make the system faster, cheaper and higher quality. See our docs on context management for more details.

Local Search Tools (RAG)

When we added RAG (Retrieval-augmented generation) last release, one of the top requests was to be able to build a RAG system completely locally, privately, without any third party services. Well, now you can! We've added a range of open-weight and local models in our new release:

• Extraction models (vision models which convert documents into text for RAG indexing): Qwen 2.5VL 3B/7B/32B/72B, Qwen 3VL and GLM 4.5 Vision
• Embedding models: Qwen 3 embedding 0.6B/4B/8B, Embed Gemma 300M, Nomic Embed 1.5, ModernBert, M2 Bert, E5, BAAI/bge, and more

Note: Some of the models above are open models, but we're still waiting on projects like Llama.cpp and Ollama to add support for running locally.

Deploying Kiln RAG into Production

We've published a new guide and code for deploying RAG Search Tools built in Kiln, into production. You have several options including our MCP server and custom LlamaIndex deployments.

Kiln v0.22.0 App Release

Our new v0.22.0 app release includes:

• Agent Builder: See above for more details.
• Local + Open RAG: See above for more details.
• Save your Run Configuration & Set Default: Our Run tab now allows you to save a specific run configuration (model, prompt, parameters). This makes it easy to jump between configs. Set your preferred config as "Default", which will load automatically.
• Tools & MCP Updates: Fixed a bug impacting some MCP servers on Windows, Better error messages when MCP servers error
• New Models: Sonnet 4.5, GLM 4.6, GLM 4.5 Vision, Mistral Medium 3.1, Qwen 3 VL, Qwen 2.5 VL
• And More: UI improvements, faster tagging on large datasets, tagging UI shows all tags you've used, simplified onboarding, new intro screen, and more

Download Kiln