ogx-ai/ogx

OGX Explainer (based solely on the README) Task OGX provides an OpenAI‑compatible API server that lets you run LLMs and agentic workflows anywhere—laptop, data‑center, or cloud. Input HTTP requests to the server’s endpoints, e.g.: * /v1/chat/completions – chat prompts * /v1/embeddings – text to embed * /v1/files – file uploads * /v1/vector_stores – documents for RAG * /v1/batches – batch job specifications * /v1alpha/skills – skill bundle manifests * Anthropic /v1/messages and Google GenAI /v1alpha/interactions calls Output Responses from the server, such as: * Generated chat completions * Embedding vectors * File and vector‑store IDs * Batch job status/results * Skill invocation outputs * Responses from Anthropic or Google GenAI APIs Architecture * Pluggable provider layer – swap back‑ends (Ollama, vLLM, managed services) without changing the API. * Core API layer – implements OpenAI Chat/Completions/Embeddings, Responses API (agentic orchestration with tool calling, MCP, file search), Vector Stores & Files, Batches, Skills, plus native Anthropic and Google GenAI SDK endpoints. * Open Responses conformance – the Responses API passes the Open Responses test suite. * Multi‑SDK support – clients can use the OpenAI, Anthropic, or Google GenAI SDKs against the same server. Training Data N/A – OGX is a serving framework, not a model that is trained. Loss N/A – no training loss is involved. How to run (from README) bash # Install curl -LsSf https://github.com/ogx-ai/ogx/raw/main/scripts/install.sh | bash # or: uv pip install ogx # Start uv run ogx go Then point any OpenAI‑compatible client (or Anthropic/Google GenAI client) at http://localhost:8321/v1 (or the appropriate SDK endpoint). Key capabilities highlighted * Drop‑in replacement for the OpenAI API. * Model‑agnostic: works with Llama, GPT, Gemini, Mistral, etc. * Built‑in agentic orchestration (Responses API) with tool calling, MCP server integration, and file‑search RAG. * Managed vector stores and file storage. * Offline batch processing via /v1/batches. * Versioned skill bundles (/v1alpha/skills). * Official Python and TypeScript client SDKs, plus an OpenAPI‑generated alternative. * Community support via Discord and regular calls.