atropos/environments/hack0/README.md
2025-05-18 13:00:43 -07:00

1.2 KiB

Readme

1-Minute Demo Video

Watch the demo on YouTube

Environment Design & Motivation

NousWhiteHouse is a reinforcement learning (RL) project focused on improving agent tool calls using the Model Context Protocol (MCP). The goal is to enable agents to dynamically discover and invoke tools more effectively, leveraging MCP for context-aware decision-making.

After replicating RESTGPT, we noticed that LLMs struggled to find the right tools to call, such as finding Gims songs on Spotify. Instead of manually matching multiple APIs, the recent advent of Modex Context Protocol inspires us to double down on tool-calling efforts

[to add] Our main task or challenge that our environment presented [to add] Why is this environment interesting or useful for RL research [to add] What inspired our design choices

🔖 Environment Snapshot

Field Entry
Environment Name
Short Description
Category
Dataset Needed?
External Deps
Environmental Variables
Compute Footprint Estimate

Estimate 🧪 Zero-Training Test Results Details W&B Link:

Examples of the Environment scoring a good example and a bad example