Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

GitHub - shobrook/saplings: Build smarter AI agents using tree search

Nov 26, 2024 - github.com
Saplings is a framework designed to build agents that use search algorithms to solve problems. It supports popular search algorithms like Monte Carlo Tree Search (MCTS), A*, and greedy best-first search, and uses OpenAI function calling. The framework is easy to use, allowing users to add search to their agent in just two lines of code. It provides full control over the evaluation function, prompts, search parameters, and more. Saplings has been shown to enhance reasoning and overall task performance compared to traditional, ReAct-style agents.

The framework includes features such as installation, quickstart, creating a tool, configuring an agent, and documentation. It supports OpenAI models and has plans to include Anthropic and Groq models. The agents in Saplings have parameters like tools, model, evaluator, prompt, branching factor, maximum depth, threshold, verbosity, tool choice, and parallel tool calls. The framework also provides advanced tool options like accessing agent memory and reformatting tool output. Saplings is on a mission to build the easiest and most powerful framework for building search-enabled agents as more inference-time compute makes agents smarter.

Key takeaways:

  • Saplings is a framework for building agents that use search algorithms to solve problems. It supports popular search algorithms like Monte Carlo Tree Search (MCTS), A*, and greedy best-first search.
  • The framework provides full control over the evaluation function, prompts, search parameters, etc. It uses OpenAI function calling under the hood and has been shown to boost reasoning and overall task performance.
  • It provides a simple way to add search to your agent with just a few lines of code. The framework also allows for the creation of custom tools and evaluators, and supports the configuration of agents.
  • Future plans for Saplings include support for chat history, Anthropic and Groq models, dynamic system prompts and tool schemas, vision agents, and the addition of an 'llm_call_budget' parameter to every agent.
View Full Article

Comments (0)

Be the first to comment!