OGX

// Overview

OGX is an open-source agentic API server for building AI applications. It is a drop-in replacement for the OpenAI API that runs on a laptop, in a datacenter or in the cloud, so any OpenAI-compatible client or agent framework can talk to it unchanged. The project is the continuation of Meta's Llama Stack: the repository states plainly that Llama Stack is now OGX, and that the mission changed with the name — model-agnostic, multi-SDK, production-grade.

The server covers the standard surface — `/v1/chat/completions`, `/v1/completions` and `/v1/embeddings` — plus the Responses API for server-side agentic orchestration, where tool calling, MCP server integration and built-in file search (RAG) happen inside a single API call. Around that sit `/v1/vector_stores` and `/v1/files` for managed document storage and search, `/v1/batches` for offline batch work, and `/v1alpha/skills` for versioned skill bundles — zip archives with a SKILL.md manifest that agents can invoke. The Responses implementation passes the Open Responses conformance suite.

Multi-SDK support is the other half of the pitch: alongside the OpenAI API, OGX natively serves the Anthropic Messages API at `/v1/messages` and the Google GenAI interactions API at `/v1alpha/interactions`. A pluggable provider architecture keeps the API stable while the backend changes — develop against Ollama locally, deploy on vLLM in production, or point at a managed service. Official client SDKs exist for Python and TypeScript, with an OpenAPI-generated alternative for teams whose tooling needs it.

// What it does

• OpenAI-compatible chat completions, completions and embeddings endpoints, usable from any OpenAI client
• Responses API with server-side tool calling, MCP server integration and built-in file search, verified against the Open Responses conformance suite
• Vector stores and files endpoints for managed document storage and retrieval
• Batches endpoint for offline, high-volume processing
• Skills — versioned zip bundles with a SKILL.md manifest that agents can call at /v1alpha/skills
• Native Anthropic Messages and Google GenAI interactions endpoints alongside the OpenAI surface
• Pluggable providers so the same API runs on Ollama locally and vLLM or a managed service in production
• Official Python and TypeScript client SDKs (ogx_client), plus an OpenAPI-generated alternative

// Getting started

OGX installs as a Python package and starts a server that auto-detects providers from your environment.

# One-line install
curl -LsSf https://github.com/ogx-ai/ogx/raw/main/scripts/install.sh | bash

# Or via uv
uv pip install ogx

uv run ogx go

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8321/v1", api_key="fake")
response = client.chat.completions.create(
    model="llama-3.3-70b",
    messages=[{"role": "user", "content": "Hello"}],
)

pip install ogx_client
# or
npm install ogx-client

Commands and code are distilled from the project's own documentation — always check the official repo for the latest.

// When to use it

• Put one OpenAI-compatible endpoint in front of several model providers so application code never changes when the model does
• Serve the Responses API self-hosted, with tool calls, MCP servers and file search resolved server-side
• Develop against Ollama on a laptop and deploy the same stack on vLLM without touching the client
• Expose the OpenAI, Anthropic and Google GenAI SDK surfaces from a single server for a mixed-SDK codebase

// How OGX compares

OGX alongside other open-source app frameworks tools AI/TLDR tracks, ranked by GitHub stars.