Toolhouse
Last updated
Last updated
Toolhouse is a tool management platform for LLM tool use (also known as function calling).
With Toolhouse, you can define tools and optimize them for each LLM you use. This allows you to use the same code across multiple LLMs without rewriting tool definitions or prompts to make them work across LLMs. Toolhouse can log tool inputs and outputs for each LLM call for easy debugging and testing.
Toolhouse is an open framework that works with any LLM that supports tool use. As most developers use more than one LLM provider in their deployments, Toolhouse is designed to help you manage your functions across each model.
Want to see how Toolhouse works? Use our Quick start.
Toolhouse helps you create and specify tool definitions in a convenient platform. Each tool's description (including descriptions for their arguments) can be edited in the platform. This way, you can allow your prompt engineering team to create and optimize tool descriptions without the need to change any code.
In our roadmap: Create prompt variations for each tool, so you can customize a tool based on its target LLM.
The Toolhouse SDK helps you pass tools (including tool choices) to your LLM, and to easily process its tool calls. The SDK streamlines the tool calling code, so you can avoid building boilerplate code. The SDK works alongside your existing LLM SDK, and it allows you to inspect each tool call in case you need to perform further processing.
Execution logs
With Execution logs, you can inspect any function called by the LLM, along with their arguments and return values. Execution logs are useful to understand the calls being made by a specific model, as well as the type of arguments and the number of tokens consumed. Each execution is timestamped so you can understand execution time and optimize latency accordingly.
You can transform APIs into a tool call with Remote Tools. With Remote Tools, you only have to specify an API definition and authentication keys. When your LLM calls a remote function, Toolhouse will perform the related API request for you and will pass the output to the LLM without the need to write your API logic.
In our roadmap: Remote tools will have the ability to cache their responses; developers will be able to specify test data and mock responses.