AgentLab and HelloAgents
AgentLab provides a comprehensive benchmarking and testing infrastructure for web agents, while HelloAgents offers a lightweight tutorial-based framework for agent development, making them complements that could be used together where HelloAgents serves as a starting point and AgentLab validates the resulting agents.
About AgentLab
ServiceNow/AgentLab
AgentLab: An open-source framework for developing, testing, and benchmarking web agents on diverse tasks, designed for scalability and reproducibility.
Built on **BrowserGym** for standardized web task environments, AgentLab provides large-scale parallel experiment execution via Ray with unified LLM API support across OpenAI, Azure, OpenRouter, and self-hosted TGI backends. It supports 12+ benchmarks including WebArena, WorkArena, and VisualWebArena with configurable reproducibility features like task seeding and deterministic execution, enabling systematic ablation studies and agent comparisons across thousands of tasks.
About HelloAgents
jjyaoao/HelloAgents
A agent framework based on the tutorial hello-agents
Implements 16 production-grade capabilities including ToolResponse protocol, context engineering (HistoryManager/TokenCounter), session persistence, sub-agent mechanisms via TaskTool, circuit breakers, and observability through TraceLogger. Built on OpenAI-compatible APIs with multi-provider support (OpenAI, Anthropic, Gemini, DeepSeek, local vLLM/Ollama) through three adapter patterns, offering Function Calling architecture across multiple agent types (SimpleAgent, ReActAgent, ReflectionAgent, PlanAndSolveAgent). Provides complete engineering infrastructure for complex multi-agent applications including streaming output (SSE), async lifecycles, optimistic locking for file operations, and decision logging.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work