OGX

OGX (formerly Llama Stack) is an open-source AI application server maintained by the ogx-ai organization. It exposes an OpenAI-compatible HTTP API that combines inference, vector stores, file storage, tool calling, and agentic orchestration, and it can also be imported as a Python library. The project's latest release listed is v1.5.0.

What It Is

OGX is an application server for building AI apps. Applications talk to a standard API over HTTP, so any language or OpenAI-compatible client can be used, while the backend providers can be swapped without changing application code. The docs describe it as a drop-in replacement for the OpenAI API that can run on a laptop, in a datacenter, or in the cloud.

APIs and Orchestration

The server provides Chat Completions, Embeddings, Models, Vector Stores, Files, and Batches endpoints. Its Responses API handles server-side agentic orchestration with tool calling, MCP integration, and built-in file search (RAG). The docs state the Responses implementation conforms to the Open Responses specification. Beyond OpenAI endpoints, it offers an Anthropic Messages adapter, a Google GenAI Interactions endpoint, Prompts, File Processors, Connectors, and an alpha Skills API for versioned skill bundles.

Providers and Deployment

A pluggable provider architecture covers inference (such as Ollama, vLLM, OpenAI, Anthropic, AWS Bedrock, Azure OpenAI, and Gemini), vector stores (such as FAISS, PGVector, Qdrant, ChromaDB, and Milvus), and tool runtimes including web search and MCP. Guardrails call an external OpenAI-compatible moderation endpoint. A typical path is developing locally with Ollama and FAISS and deploying with vLLM and PGVector. The docs also describe pointing Claude Code, Codex, or OpenCode at OGX.

Setup Path

Install with a one-line script or via pip/uv, then start the server with uv run ogx go. Official Python and TypeScript client SDKs are available.