ToolProduction ReadyA2AMCPAG-UI

atmosphere

Name: atmosphere
Rating: 3.4 (3815 reviews)
Author: Atmosphere

by Atmosphere

Portable JVM agent runtime with multi-protocol support and governance

Java

Updated Sep 21, 2026

3.8k

Stars

763

Forks

Commits/Month

View on GitHub

Overview

Provides a portable JVM runtime for building and running AI agents across multiple Java frameworks. Implements a single @Agent SPI that can run on Spring AI, LangChain4j, Anthropic and more, while supporting token streaming, tool calls, human approvals and governance over WebSocket, SSE, gRPC or WebTransport/HTTP3. Ships built-in support for MCP Model Context Protocol (MCP), A2A Protocol Pattern and AG-UI protocols so agents can interoperate across networks and UIs.

Need implementation help?

Expert guidance available

The Value Proposition

As agent ecosystems become heterogeneous, interoperability and consistent runtime behavior are essential to evaluate agent interactions and track reliability. Atmosphere makes it practical to run the same agent code across different JVM stacks and protocols, which simplifies producing comparable interaction logs and trust signals. That consistency is critical for agent-to-agent evaluation, continuous evaluation pipelines, and building reputational histories across deployments. The approach also aligns with the Semantic Capability Matching Pattern to ensure capabilities are matched across diverse environments.

Ideal For

Java teams building production multi-agent systems that need consistent runtime behavior, protocol interoperability, and governance features.

How It's Used

Run identical agent code across Spring AI, LangChain4j, and Anthropic backends for apples-to-apples evaluation
Stream tokens and tool calls while capturing interaction logs for building agent track records
Bridge agents across A2A and MCP networks and expose AG-UI for consistent human-in-the-loop approvals
Standardize governance and observability for production JVM-based agents