OpenAI released Symphony on April 27, a Draft v1 open-source specification for Codex orchestration that the company describes as language-agnostic. According to a search-retrieved excerpt from openai.com, the document is framed as a spec for Codex orchestration specifically, not the broader universal multi-agent standard that some early reporting implied.
That distinction matters. A Codex orchestration spec and a general cross- architecture multi-agent protocol are different things. One governs how Codex- based agents hand off tasks, negotiate state, and share context. The other would define how agents built on different underlying models communicate across organizational and vendor boundaries. Symphony, as confirmed so far, is the former. Whether the full spec document extends to the latter remains unconfirmed at publication time.
What’s actually confirmed: Symphony carries the “language-agnostic” designation, meaning the protocol isn’t tied to a specific programming language or runtime environment. The document is at Draft v1 status, which puts it in an early, open- for-comment phase rather than a production-ready standard. The spec was published as open-source, which means other developers and organizations can review, implement, and propose amendments.
One claim from early reporting, that Symphony includes native WebSocket support via OpenAI’s Responses API to reduce latency in agentic loops, could not be independently confirmed before publication. Treat this detail as unverified until the primary spec document is reviewed directly.
Symphony enters a crowded standards space. Google’s Agent-to-Agent (A2A) protocol and Anthropic-adjacent Model Context Protocol (MCP) are both active efforts at solving similar problems: how agents built by different teams, running on different models, pass context and coordinate work without brittle custom integrations. The four agentic framework releases catalogued here on April 27 established that the orchestration stack is one of the most actively contested layers in AI infrastructure right now. Symphony is the fifth named development in ten days.
For developers, the practical question isn’t which spec wins a standards vote. It’s which protocol their tooling already supports, which one their enterprise customers will require, and whether a spec backed by OpenAI’s distribution and Codex adoption will converge with or compete against the others. Protocol fragmentation is a real cost. Each additional standard means another adapter layer, another maintenance surface, another decision before a team can ship.
The governance question runs alongside the technical one. Symphony is open-source, but OpenAI controls the initial draft. Version governance, who accepts pull requests, who sets the roadmap, who adjudicates breaking changes, will determine whether this becomes a genuinely community-governed standard or a vendor-owned spec that carries an open-source label. That question isn’t answered in Draft v1.
Independent evaluation of Symphony’s claims, scope, interoperability, memory management, is pending. Practitioners should review the spec document directly before making architectural commitments.