# waybarrios/vllm-mlx

High-performance OpenAI and Anthropic compatible LLM inference server for Apple Silicon. Native MLX, continuous batching, multimodal models, MCP tool calling, and Claude Code support.

Repository: https://github.com/waybarrios/vllm-mlx
Canonical: https://ross.abutalabs.com/products/vllm-mlx
Homepage: https://vllm-mlx.is-a.dev/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: apple-silicon, llm, macos, mlx, multimodal-ai, speech-to-text, text-to-speech, vision-language-model, vllm, anthropic, claude-code, anthropic-api, continuous-batching, inference-server, local-llm, mcp, openai, openai-api, openai-compatible, tool-calling
Last push: 2026-08-26T18:49:59+00:00

## Health v2 (maintenance only)
Score: 82/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 97, longevity 19
- inputs: {"age_days": 270, "days_push": 7, "days_rel": 21, "gap_med": 10.5, "n_releases_24m": 13}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1547, forks 216 (observed 2026-08-28T04:05:01.579840+00:00)

## What it is
vllm-mlx is a vLLM-style LLM inference server for Apple Silicon Macs built on native MLX, exposing both OpenAI and Anthropic compatible APIs from a single process. It supports continuous batching, paged and SSD-tiered KV caching, multimodal models (vision, audio, TTS, STT), MCP tool calling, and structured output.

## Use cases
- run local llm server on mac with openai api
- serve anthropic-compatible api for claude code locally
- self-host llm inference on apple silicon
- run vision language models to analyze images locally
- local speech-to-text and text-to-speech server
- batch concurrent llm requests with continuous batching
- use mcp tool calling with local models
- generate embeddings and rerank documents locally

## When to choose
- you have an Apple Silicon Mac and want high-throughput local inference without model conversion
- you need both OpenAI and Anthropic API compatibility from one server
- you want multimodal serving (text, image, audio, TTS/STT) in a single process
- you want to point Claude Code or OpenAI SDKs at a local endpoint

## When to avoid
- you need GPU inference on Linux or NVIDIA CUDA hardware
- you are on Windows or Intel Macs
- you need a battle-tested production server at massive scale
- you only need simple single-user local chat without API compatibility

## Facets
- artifact type: service
- maturity: active
- function: llm-inference, http-server, api-framework, speech-recognition, tts, mcp, chatbot, rag
- domain: large-language-models, artificial-intelligence, machine-learning, speech-processing, self-hosted, apis
- platform: python, cli, self-hosted
- tags: apple-silicon, mlx, openai-compatible, anthropic-compatible, continuous-batching, local-llm, inference-server, multimodal, claude-code, kv-cache, macos

## Member repositories
- waybarrios/vllm-mlx (main) score 82

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:01.579840+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T04:30:31.197085+00:00, confidence not recorded.
  - readme: https://github.com/waybarrios/vllm-mlx (fetched 2026-08-28T04:05:01.579840+00:00, sha 81871c841a74)
  - homepage: https://vllm-mlx.is-a.dev/ (fetched 2026-08-29T11:31:39.746815+00:00, sha ee9161916e4f)
  - site_page: https://vllm-mlx.is-a.dev/getting-started/installation (fetched 2026-08-29T11:31:39.756356+00:00, sha d78590701874)
  - site_page: https://vllm-mlx.is-a.dev/getting-started/quickstart (fetched 2026-08-29T11:31:39.759528+00:00, sha 79a1cd4a0b66)
  - site_page: https://vllm-mlx.is-a.dev/development/documentation (fetched 2026-08-29T11:31:39.762177+00:00, sha 7a2da837c1cf)
- Data as of 2026-08-30T08:39:29.467469+00:00.