Amy WorkspAIce 开源 AI 平台 v0.1：自托管、OpenAI 兼容 API

A self-hosted, customizable AI platform. It runs models on your own GPU, serves them through an OpenAI-compatible API, and includes a web chat UI.

It is the first app of Amy WorkspAIce, an open-source workplace suite (documents, spreadsheets, tasks, notes, code, chat, ...). Each app is its own container. The AI Platform is the AI service your other apps call, and it can act on those apps through their APIs.

Status: early development (v0.1). Working today: chat with streaming, a multi-provider OpenAI-compatible API, bundled Ollama with NVIDIA GPU support, API-key auth. See Roadmap.

Features

One container, zero config. Bundled Ollama with CUDA. Start it, pull a model, chat.
OpenAI-compatible API at /v1. Point any OpenAI SDK or app at it.
Bring any engine. Bundled Ollama, an external Ollama, llama.cpp server, vLLM, LM Studio, or hosted APIs. Models from all of them appear in one list, as provider/model.
Secure by default. API key required, auto-generated on first start.
Runs anywhere. Unraid template, Docker Compose, Docker Desktop (Windows/macOS), bare metal.

Quick start

Unraid

Install from Community Applications (pending submission), or add the template manually. See docs/install/unraid.md.

Docker (Linux, NVIDIA GPU)

docker run -d --name amy-ai-platform --gpus all \
  -p 8484:8484 -v ./amy-data:/data \
  ghcr.io/amy-workspaice/ai-platform:latest

docker logs amy-ai-platform          # shows your API key
docker exec -it amy-ai-platform ollama pull llama3.2

Then open http://localhost:8484. No GPU? Leave out --gpus all.

More options (compose, external Ollama, Windows/macOS): docs/install/docker.md and docs/install/other.md.

Use the API

from openai import OpenAI

client = OpenAI(base_url="http://localhost:8484/v1", api_key="amy-...")
reply = client.chat.completions.create(
    model="ollama/llama3.2",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(reply.choices[0].message.content)

Image variants

Tag	Contents	Use when
`latest`	Platform + bundled Ollama + CUDA runtime (large)	Plug-and-play, especially on Unraid
`slim`	Platform only (small)	You already run Ollama, vLLM, etc. elsewhere

Documentation

Roadmap

OpenAI-compatible proxy over multiple providers
Web chat with streaming
Bundled Ollama + NVIDIA GPU, Unraid template
Persistent conversation history
Provider and model management in the UI (pull models, VRAM-aware suggestions)
Multiple API keys and users
Customization: personas/system prompts, themes, plugins
Connectors: use other Amy WorkspAIce apps (and any MCP server) as tools
Embeddings API, RAG

Contributing

See CONTRIBUTING.md.

License

AGPL-3.0-or-later. You may self-host, modify, and redistribute it. If you offer a modified version to users over a network, you must publish your changes.