OGX (Open GenAI Stack) is an open-source agentic API server — the project formerly known as Meta’s Llama Stack, relaunched with a model-agnostic mission. It is a drop-in replacement for the OpenAI API that you can run on a laptop, in your datacenter or in the cloud, and it has about 8,400 stars on GitHub with active development in 2026.
OGX exposes standard OpenAI-compatible endpoints — chat completions, embeddings and the Responses API — while letting you swap the underlying model freely between Llama, GPT, Gemini, Mistral and many others without changing application code. Its Responses API implementation adds server-side agentic orchestration with tool calling, MCP server integration and built-in file search (RAG) in a single call, and it also provides vector stores, file management, batch processing and versioned agent “skills”. A pluggable provider architecture means you can develop locally with Ollama and deploy to production on vLLM or a managed service behind the same API, and it natively accepts OpenAI, Anthropic and Google GenAI SDK clients. The project descends directly from the Llama Stack codebase Meta launched in 2024, and keeps its distribution model: pre-configured stacks that bundle an inference provider, vector store and agent runtime for common setups, from a developer laptop to a Kubernetes cluster. That heritage makes OGX a natural migration path for existing Llama Stack users, and its conformance testing against the Open Responses specification aims to keep every provider behind exactly the same observable API behavior.
OGX is completely free and open source under the MIT license. You only pay for the models and infrastructure you connect it to.
OGX is for teams who want one stable, standards-based API layer for AI applications — portable across models and infrastructure — instead of being locked into a single vendor’s API.