atmosphere

View on GitHub

Portable AI agent runtime for the JVM. One @Agent class runs on Spring AI, LangChain4j, Anthropic, or 9 more behind one SPI. Token streaming, tool calls, human approvals, and governance over WebSocket, SSE, gRPC, or WebTransport/HTTP3. Speaks MCP, A2A, and AG-UI.

★ 3.8K763 forksJavaApache-2.0Atmosphere

Atmosphere is a JVM runtime for stateful AI agents: one @Agent class streams tokens and tool calls over WebSocket/SSE/gRPC and runs on Spring AI, LangChain4j, Anthropic, or other adapters via a single AgentRuntime SPI. It packages multi-agent coordination, MCP/A2A/AG-UI endpoints, human approvals, memory, and governanc

Use Cases

Declare a deep agent with a single @Agent annotationStream LLM tokens to browsers over WebSocket/SSE/gRPCMulti-agent coordination with @Coordinator and AgentFleetHuman-in-the-loop approval and durable workflowsExpose agents over MCP, A2A, and AG-UIGovern tool calls with policy admission, cost ceilings, PII redactionChat bots for Slack, Telegram, Discord, WhatsApp, MessengerSession tape replay for debugging and distillationSwap LLM runtime adapters without rewriting endpointsLong-term memory plus vector-store semantic recall

Built With

Language: Java
Frameworks: Spring AI · LangChain4j · Google ADK · Embabel · Koog · Semantic Kernel · AgentScope · Spring AI Alibaba · CrewAI · Quarkus · Spring Boot · Netty · Jetty · Undertow

atmosphere

Use Cases

Built With

Tags