atmosphere

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

What is atmosphere?

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.

Overview of atmosphere

What it is

Atmosphere is a portable AI agent runtime for the JVM. A single @Agent class can run on Spring AI, LangChain4j, Anthropic, or nine more AI frameworks behind one Service Provider Interface (SPI). It provides token streaming, tool calls, human approvals, and governance over WebSocket, SSE, gRPC, or WebTransport/HTTP3. Atmosphere speaks the MCP, A2A, and AG-UI protocols. It is designed for teams that need AI agents to behave like production services: streaming over real transports, guarded before every tool call, observable by tenant and run, and portable across AI frameworks without rewriting the endpoint.

Key features

  • Streaming to real clients via WebSocket, SSE, long-polling, and gRPC as always-on defaults; WebTransport over HTTP/3 is optional.
  • One AgentRuntime SPI with twelve runtime adapters and contract-tested capability flags for swapping AI integrations.
  • A plain @Agent is batteries-included with memory, a plan (write_todos), a virtual filesystem, and sub-agent delegation (task).
  • Governance features: policy admission, @AgentScope, human approval, plan-and-verify, cost ceilings, PII rewriting, and admin kill switches.
  • Durable human-in-the-loop (HITL) approvals that hibernate without holding a thread, persist workflow state, and resume through REST approval surfaces.
  • Durable sessions, run IDs, replay buffers, checkpoints, and reconnect-safe continuation.
  • Opt-in session tape recording session-boundary AI events (SQLite-durable) for reconstruction and distillation.
  • Browser endpoints plus MCP (stateless RC on the MCP 2026-07-28 spec, sessions back to 2024-11-05), A2A, AG-UI, Slack, Telegram, Discord, WhatsApp, and Messenger modules.

Use cases

Atmosphere is built for human-in-the-loop, multi-channel, governed agents inside an existing JVM stack. It is suitable for production services that require real-time streaming, policy enforcement, observability by tenant and run, and portability across AI frameworks. For stateless, bursty, autonomous agents that should hibernate when idle, a serverless agent platform is usually a better fit.

Getting started

brew install Atmosphere/tap/atmosphere# Or:curl -fsSL https://raw.githubusercontent.com/Atmosphere/atmosphere/main/cli/install.sh | shatmosphere run spring-boot-multi-agent-startup-teamatmosphere new my-agent --template ai-chatcd my-agentLLM_API_KEY=your-key ./mvnw spring-boot:runatmosphere new my-agent --template ai-tools --runtime langchain4j --forceatmosphere import https://github.com/anthropics/skills/blob/main/skills/frontend-design/SKILL.md

License and pricing

Atmosphere is open source under the Apache-2.0 license.

Based on the project's public README and repository metadata.

Add this badge to your README

Featured on YunoDB Badge styles & help

Maintain this project? Show that it is featured on YunoDB.

Markdown[![Featured on YunoDB](https://yunodb.com/badge/atmosphereatmosphere.svg)](https://yunodb.com/tools/atmosphereatmosphere?utm_source=github&utm_medium=badge)
HTML<a href="https://yunodb.com/tools/atmosphereatmosphere?utm_source=github&utm_medium=badge"><img src="https://yunodb.com/badge/atmosphereatmosphere.svg" alt="Featured on YunoDB" height="20"></a>