Agent frameworks & SDKs
envharness
google-research/envharness
A research framework that wraps static benchmark environments with composable Setup, Rule and Link layers so agents can learn, induce skills and train with RL across ALFWorld, WebArena, SWE-bench and more.
Related projects
autogen61.8kA framework for building multi-agent AI applications with layered runtimes, orchestration, and extensible tooling.
crewAI60.7kA Python framework for building role-based multi-agent crews and event-driven flows for production automation.
private-gpt57.6kAn open-source API layer over local models that adds retrieval, tools, and MCP so developers can build private AI apps.
agno43.1kA Python framework and platform for building, running and improving agents, with teams, workflows, memory, tools, durable execution, APIs, MCP servers and observability.
dspy39kA declarative framework for programming foundation models, optimizing prompts for RAG pipelines and agent loops.
CopilotKit38.7kAn SDK and interaction layer for building agent-native apps with chat, generative UI, shared state, and human-in-the-loop across frameworks.