Back to directory

Atmosphere

A portable, real-time runtime for AI agents on the JVM that streams a single agent implementation to clients over WebSocket, SSE, gRPC or WebTransport, and speaks MCP, A2A and AG-UI.

Atmosphere

Atmosphere is a real-time, event-driven framework for running AI agents as production JVM services. Tokens generated by an agent flow from the model runtime to a client through a broadcaster pipeline that a developer can filter, gate and observe, reaching clients over WebSocket, SSE, long polling, gRPC or, optionally, WebTransport over HTTP/3, and reaching other systems through MCP, A2A and AG-UI.

A single agent implementation is portable across twelve supported runtime adapters, such as Spring AI, LangChain4j and a direct Anthropic integration, behind one service-provider interface with contract-tested capability flags. Batteries-included features include memory, a plan and to-do mechanism, a virtual filesystem, sub-agent delegation, policy admission checks, human-in-the-loop approvals that persist without holding a thread, durable sessions with replay buffers and checkpoints, and an opt-in session tape for replaying or distilling a run.

It runs inside any servlet container, such as Tomcat, Jetty, Netty, Undertow, Quarkus or Spring Boot, which owns compute and scheduling; Atmosphere does not host agents or handle payments itself.

More in Agent frameworks

See all 902
  • Recurse

    recurse.run

    A serverless harness that lets a coding agent design, test, and deploy its own specialist sub-agents as tools, MCP servers, or bots, defined by a schema-validated manifest.

  • ProofForge

    shulgin.is-a.dev

    An AI-agent pipeline that decomposes a mathematics problem, proves each piece, and formalizes the result in Lean 4 and Mathlib until the Lean kernel accepts it, with six of its proofs already merged into Google DeepMind's formal-conjectures repository.

  • local-coder

    github.com

    Builds a local coding-agent team on top of Ollama and OpenCode by detecting the machine's hardware, recommending models for seven specialist roles, and testing whether each assigned model can actually use tools, edit a repository and pass verification.