What changed? OpenAI donated the Open Responses API specification to the Agentic AI Foundation (AAIF), a neutral home hosted by the Linux Foundation. The spec defines a set of core calls—messages, tool calls, function calling—that aim to work uniformly across LLM providers. Existing implementations such as Vercel AI Gateway and Databricks Model Serving already support the spec.
Standardizing LLM Primitives
The Open Responses spec abstracts the common building blocks that most LLM services expose. By codifying these primitives, the specification removes the need for developers to write provider‑specific adapters when swapping models. The core schema is derived from OpenAI’s own API, reflecting the de‑facto patterns that have emerged from earlier Chat Completions and Anthropic’s Messages APIs.
Adoption Landscape
Several major platforms have announced support for the spec, including:
- AWS (Bedrock)
- Databricks
- LM Studio
- Ollama
- Hugging Face
- Nvidia
- OpenRouter
- Red Hat
- Vercel (via AI Gateway)
- Open‑source runtimes such as
vLLMand OGC (formerly Llama Stack)
Notably, Anthropic and Google are not listed as current adopters. Anthropic runs its own Messages API, while Google offers an Interactions API. Both are members of AAIF, so future alignment is possible but not guaranteed.
Operational and Security Considerations
From an operational standpoint, the spec simplifies CI/CD pipelines that target multiple LLM back‑ends. Teams can standardize on a single request format, reducing the surface area for configuration drift. Existing gateway services that already implement the spec can be leveraged as drop‑in layers, potentially lowering latency and cost of integration.
Security implications are indirect. A common request schema can make audit logging and traffic inspection more consistent across providers, but the spec itself does not prescribe authentication or authorization mechanisms. Practitioners should still verify that each provider’s endpoint enforces its own identity and access controls, and that any gateway (e.g., Vercel AI Gateway) correctly propagates credentials.
Related CloudNinjas coverage: AI engineering.
What This Means For Practitioners
Engineers should evaluate whether their current LLM integration layer already aligns with the Open Responses primitives. If not, consider refactoring to the spec to gain:
- Reduced code duplication when supporting multiple vendors.
- Easier migration between models without rewriting request handling.
- Access to community‑driven extensions via AAIF’s neutral governance.
Watch for upcoming announcements from Anthropic and Google, as their participation could further consolidate the ecosystem. In the meantime, validate that any gateway or custom wrapper you use adheres to the spec’s schema and that you retain visibility into provider‑specific security controls.