OGX
MetaMeta's open inference + deployment runtime for LLMs; originally launched as Llama Stack (`meta-llama/llama-stack`) in 2024 and rebranded in 2026 as OGX / Open GenAI Stack (`ogx-ai/ogx`) with model-agnostic positioning. Unified APIs for inference, agents, fine-tuning, and eval; 25+ ecosystem partners including AWS, NVIDIA, Databricks, Groq, and Azure. MIT licensed, 8.4K+ stars with continuous commit history across the rename. Anchors the 1B+ Llama downloads ecosystem.
OGX (Open GenAI Stack), formerly Meta's Llama Stack; renamed/rescoped to a model-agnostic, multi-SDK agentic API server. v1.0.2 (May 13, 2026), MIT-licensed, ~8.4k GitHub stars. NOT itself an inference engine: it composes/routes to inference backends (vLLM, Ollama, Bedrock) plus vector stores, safety, tools, file storage.
Openness
5 high confidence- license
- MIT(OSI)
- source
- public
- core-gated
- ungated
Fully open (MIT, full source). Openness scored straightforwardly; the issue is category fit, not openness.
- https://github.com/ogx-ai/ogx recorded 2026-06-04
MIT license; v1.0.2 (May 13 2026); ~8.4k stars; agentic API server composing inference providers + vector stores + tools
- https://ogx-ai.github.io/blog/from-llama-stack-to-ogx recorded 2026-06-04
renamed from Llama Stack; model-agnostic, multi-SDK, routes to inference backends rather than serving weights itself
Adoption
2 low confidenceOnly ~8.4k GitHub stars and no published download/usage volume found this run; capped low on stars fallback (cannot exceed level 3). Recently renamed from Llama Stack, so historical signal is muddled.
- https://github.com/ogx-ai/ogx recorded 2026-06-04
~8.4k stars; no download/usage figure published
Capability
3 medium confidenceCapability is gateway/agentic-server breadth (multi-API-surface, provider composition), not inference throughput. As an inference engine it scores low because it does no forward pass itself; scored on its actual feature set in its true category. See recategorize.
- https://github.com/ogx-ai/ogx recorded 2026-06-04
composes 23 inference providers + vector/safety/tool/file backends behind OpenAI/Anthropic-compatible API
Unchanged since 2026-07-30 (last edited, not re-checked)