zmedelis/bosquet

Tooling to build LLM applications: prompt templating and composition, agents, LLM memory, and other instruments for builders of AI applications.

What it solves

Bosquet simplifies the development of complex AI applications by providing a structured framework for managing prompt templates, LLM context windows, and agentic interactions with external tools and APIs.

How it works

Bosquet acts as an LLMOps layer that integrates several key components:

  • Prompt Management: Uses the Selmer templating library to handle dynamic prompt creation.
  • Execution Graph: Employs a Pathom graph processing machine to enable prompt chaining and composition.
  • Agentic Tools: Provides abstractions for defining tools that LLMs (such as those from OpenAI and Ollama) can call to interact with external data or perform calculations.
  • Service Management: Handles LLM service configuration, API keys, and response caching to optimize performance and cost.
  • Memory Handling: Manages LLM memory to work around limited context windows.

Who it’s for

Developers building non-trivial AI applications who need a way to organize complex prompt workflows, integrate external tools, and manage the lifecycle of LLM interactions.

Highlights

  • Prompt Chaining: Ability to compose multiple prompts where the output of one serves as the input to another.
  • Tool Integration: Built-in support for defining and executing tools for OpenAI and Ollama models.
  • Flexible Templating: Integration with Selmer for sophisticated prompt templates.
  • CLI Support: Includes a command-line interface for testing prompts and managing LLM settings.