Repolode
Submit
Atmosphere

Atmosphere/atmosphere

A real-time runtime for JVM AI agents that streams tokens over WebSocket, SSE, or gRPC, with one SPI across 12 LLM framework adapters.

View on GitHubWebsite
Stars
3,830
Forks
763
Open issues
5
Last commit
yesterday

What is it?

Atmosphere is a Java framework for running AI agents as real-time services. A plain @Agent class gets a browser endpoint with token streaming, commands, tool calls, and memory, and the same agent can be exposed over MCP, A2A, and AG-UI. Transports include WebSocket, SSE, long-polling and gRPC, with WebTransport over HTTP/3 as an optional add-on.

The runtime sits behind an AgentRuntime SPI with adapters for Spring AI, LangChain4j, Google ADK, Embabel, Koog, Semantic Kernel, AgentScope, Anthropic, Cohere, CrewAI and others, so application code stays the same when the underlying framework changes. The project is aimed at JVM teams building human-in-the-loop or multi-channel agents that need governance. It covers policy admission, scoped prompts, tool approval, PII redaction, cost ceilings, durable checkpoints and resumable sessions.

The project also ships Slack, Telegram, Discord, WhatsApp and Messenger channel modules, a CLI for scaffolding apps, and a TypeScript client (atmosphere.js) for React, Vue, Svelte and React Native.

Key features

  • 01Streams agent tokens to clients over WebSocket, SSE, long-polling and gRPC
  • 02One AgentRuntime SPI with adapters for Spring AI, LangChain4j, ADK and other frameworks
  • 03Exposes the same agent via MCP, A2A and AG-UI endpoints
  • 04Policy admission, tool approval, PII redaction and per-tenant cost ceilings
  • 05Durable HITL approvals and resumable sessions with checkpoints
  • 06Channel adapters for Slack, Telegram, Discord, WhatsApp and Messenger

Use cases

  • Adding a streaming AI chat endpoint to an existing Spring Boot or Quarkus service
  • Building governed agents that need human approval before running tools
  • Exposing one agent to browsers, MCP clients and other A2A agents at once
  • Running a multi-channel support bot across Slack and Telegram from a single JVM codebase

Quick start

  1. 1

    Install the Atmosphere CLI with Homebrew

    brew install Atmosphere/tap/atmosphere
  2. 2

    Run a sample app with the CLI

    atmosphere run spring-boot-multi-agent-startup-team

Things to know

  • Requires an LLM API key for the default AI features.

In collections

Open-source AI agent frameworks & appsMCP serversStreaming tools

Topics

#ai-agents#java#llm#streaming#websocket#mcp#spring-boot#agentic-ai#a2a#acp

FAQ

What is atmosphere?
A real-time runtime for JVM AI agents that streams tokens over WebSocket, SSE, or gRPC, with one SPI across 12 LLM framework adapters. Atmosphere is a Java framework for running AI agents as real-time services. A plain @Agent class gets a browser endpoint with token streaming, commands, tool calls, and memory, and the same agent can be exposed over MCP, A2A, and AG-UI. Transports include WebSocket, SSE, long-polling and gRPC, with WebTransport over HTTP/3 as an optional add-on.
Is atmosphere free to use? What license does it have?
atmosphere is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is atmosphere still maintained?
Actively maintained. The last commit to atmosphere was on Nov 6, 2026.
What can I use atmosphere for?
Adding a streaming AI chat endpoint to an existing Spring Boot or Quarkus service Building governed agents that need human approval before running tools Exposing one agent to browsers, MCP clients and other A2A agents at once Running a multi-channel support bot across Slack and Telegram from a single JVM codebase
How do I install atmosphere?
Install the Atmosphere CLI with Homebrew: brew install Atmosphere/tap/atmosphere Run a sample app with the CLI: atmosphere run spring-boot-multi-agent-startup-team
Where can I get atmosphere?
The source code and installation instructions are on GitHub: https://github.com/Atmosphere/atmosphere

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects