Open Responses: explicit spec for OpenAI's Responses API supported by OpenRouter, Ollama, Huggingface, vLLM, et al

**OpenAI** launched the **Open Responses** API spec, an open-source, multi-provider standard for interoperable LLM APIs designed to simplify agent stacks and tooling. Early adopters like **ollama** and **vLLM** support the spec, while notable absences include **anthropic** and **google-deepmind**. Agent design insights from **Cursor** emphasize explicit roles and planning over mega-agent models, with **GPT-5.2** outperforming **Opus 4.5** in long runs. The emerging dominant context/memory abstraction for agents is a **filesystem-as-memory** approach, championed by **llamaindex** and **langchain**, using virtual filesystems often backed by databases like Postgres. LangChain also shipped an open-source desktop interface for agent orchestration called **openwork**. This news highlights advances in API standardization, agent architecture, and memory abstractions in AI development.