Atmosphere
ActiveDescription
Real-time transport layer for Java AI agents supporting WebSocket, SSE, gRPC, and WebTransport/HTTP3, with native MCP, A2A, and AG-UI protocol support for building event-driven AI agent communication architectures.
Key Features
- JVM runtime for streaming AI agents with @Agent annotation-driven development — one annotation declares the agent, classpath modules decide integrations
- Real-time transport layer supporting WebSocket, SSE, long-polling (always-on defaults), gRPC, and optional WebTransport/HTTP/3
- 12 runtime adapters with contract-tested capability flags — swap between Spring AI, LangChain4j, Google ADK, Anthropic, Cohere, CrewAI and more
- Production governance with policy admission, @AgentScope, human-in-the-loop approvals, plan-and-verify, PII redaction, and admin kill switches
- Durable sessions with run IDs, replay buffers, checkpoints, and reconnect-safe continuation for long-running agent workflows
- Multi-channel protocol surface — MCP, A2A, AG-UI, plus Slack, Telegram, Discord, WhatsApp, and Messenger channel adapters
Use Cases
💡 Building production AI agents that need real-time streaming transport with governance and policy enforcement
💡 Enterprise JVM applications requiring agent orchestration with human-in-the-loop approval workflows
💡 Multi-channel AI deployments serving browser, Slack, Telegram, Discord, and WhatsApp from a single agent definition
💡 Regulated industries needing PII redaction, cost ceilings, and admin kill switches for AI agent operations
💡 Teams porting existing Java/Spring applications that want to add AI agent capabilities without rewriting endpoints
Categories
Quick Start
Install via `brew install Atmosphere/tap/atmosphere` or the curl installer. Run a sample: `atmosphere run spring-boot-multi-agent-startup-team`. Create your own: `atmosphere new my-agent --template ai-chat`, then set `LLM_API_KEY` and run `./mvnw spring-boot:run`.