Crab
Log In
CRAB is an open‑source, Python-first framework for building and benchmarking multimodal LLM agents that operate across multiple environments—desktop, mobile, Docker, or cloud VM.
Overview
Cross‑environment orchestration: Runs agents concurrently on desktop, mobile, Docker, and cloud
Python-native environment definition: Use decorators to define environment actions
Graph-based benchmarking: Fine-grained evaluation using CRAB Benchmark‑v0 (120 tasks across Ubuntu & Android)
Tool-agnostic: Works with any Python-accessible environment via API wrappers
Demo Screens
Capabilities
Environment Setup & Control
Launches and controls tasks across multiple environments—e.g., spinning up VMs or mobile emulators.
Task Benchmarking
Designs and evaluates agent behaviors with structured success metrics and graph evaluators.
Multi-Agent Coordination
Enables agents to coordinate across environments, using graph protocols for communication.
Scout Summary
Rating
No reviews yet
Log In
Details
Creator
Camel-AI community
Type
Externally Hosted Agent