OGX: An Open-Source, Vendor-Neutral Generative AI Application Server
| Source: arXiv AI
Tags: OGX, open-source, vendor lock-in, inference server, OpenAI API, Anthropic, agentic AI
OGX (Open GenAI Stack) is a mature open-source AI application server with 8,400+ GitHub stars that implements OpenAI, Anthropic, and Google APIs with pluggable backends — letting developers code against a single API surface and switch among 20+ inference providers and 13 vector store backends without changing application code.
Details
OGX addresses a real friction point for teams building production AI applications: the proliferation of incompatible APIs across providers. By implementing the OpenAI Responses API, Anthropic Messages API, and Google GenAI Interactions API in a single server, OGX lets development teams write code once against their preferred API surface and switch inference providers without application changes. The project supports 20+ inference providers and 13 vector store backends, with a companion Kubernetes Operator for production deployment. Its primary focus is the Responses API for server-side agentic orchestration, conforming to the Open Responses specification. OGX serves as the self-hosted backend for several AI coding tools including Claude Code, Codex CLI, OpenCode, and OpenHands. With 8,400 GitHub stars, 242 contributors, and 4,000 commits across nearly two years of development, this is not a new proof-of-concept but a mature open-source project receiving a formal arXiv writeup. For enterprises concerned about vendor lock-in, OGX offers a concrete alternative to relying solely on cloud provider SDKs, enabling air-gapped or on-premises deployments. The arXiv paper is a technical description rather than an evaluation with novel benchmarks. Claims about serving Claude Code and other tools as backends are made but not independently benchmarked in the paper itself.