项目文件夹

Each extension is a self-contained directory with a manifest.yaml (service metadata), compose.yaml (Docker Compose fragment), and optional Dockerfiles, workflows, and documentation. Drop any of these into your ODS's extensions/services/ directory and run ods enable <service-id>.

Services by Category

LLM Inference & Chat

Service	Description	GPU
`ollama/`	Ollama — pull-and-run LLM server	AMD, NVIDIA
`localai/`	LocalAI — OpenAI-compatible API for local models	AMD, NVIDIA
`text-generation-webui/`	Oobabooga — full-featured LLM UI with LoRA, GPTQ, quantization	AMD, NVIDIA
`jan/`	Jan — local ChatGPT alternative with model management	AMD, NVIDIA
`librechat/`	LibreChat — multi-provider chat UI (local + cloud)	AMD, NVIDIA, Apple

Voice & Audio

Service	Description	GPU
`bark/`	Bark TTS — expressive speech with laughter, emotion, 13 languages	NVIDIA
`xtts/`	Coqui XTTS — voice cloning and multilingual TTS	AMD, NVIDIA
`piper-audio/`	Piper — fast, lightweight TTS for edge devices	AMD, NVIDIA, Apple
`rvc/`	RVC — real-time voice conversion/cloning	AMD, NVIDIA
`audiocraft/`	Meta AudioCraft — text-to-music and sound effects	NVIDIA

Image Generation

Service	Description	GPU
`comfyui/`	ComfyUI — node-based Stable Diffusion workflows	AMD, NVIDIA
`fooocus/`	Fooocus — simplified Stable Diffusion (Midjourney-like UX)	NVIDIA
`invokeai/`	InvokeAI — professional Stable Diffusion with canvas	AMD, NVIDIA
`forge/`	Forge / A1111 — Stable Diffusion WebUI with optimizations	NVIDIA

AI Development & Agents

Service	Description	GPU
`aider/`	Aider — AI pair programming in your terminal	AMD, NVIDIA, Apple
`continue/`	Continue — AI coding assistant (VS Code / JetBrains)	AMD, NVIDIA, Apple
`crewai/`	CrewAI — multi-agent orchestration framework	CPU
`gaia/`	AMD GAIA — experimental local agent UI and framework	CPU
`open-interpreter/`	Open Interpreter — natural language → system commands	CPU
`jupyter/`	Jupyter — notebooks with local LLM kernel	AMD, NVIDIA

Vector Databases

Service	Description	GPU
`chromadb/`	ChromaDB — lightweight embedding database	AMD, NVIDIA, Apple
`milvus/`	Milvus — production-grade vector database	CPU
`weaviate/`	Weaviate — vector search with hybrid ranking	CPU

Workflow Automation

Service	Description	GPU
`flowise/`	Flowise — drag-and-drop LLM chain builder	AMD, NVIDIA, Apple
`langflow/`	Langflow — visual LangChain builder	AMD, NVIDIA, Apple
`dify/`	Dify — LLMOps platform with RAG and agents	AMD, NVIDIA

Self-Hosted Apps

Service	Description	GPU
`immich/`	Immich — Google Photos alternative with AI face/object detection	AMD, NVIDIA
`paperless-ngx/`	Paperless-ngx — document management with OCR	CPU
`frigate/`	Frigate — NVR with real-time AI object detection	NVIDIA
`gitea/`	Gitea — lightweight self-hosted Git	CPU
`baserow/`	Baserow — open-source Airtable alternative	CPU
`sillytavern/`	SillyTavern — advanced roleplay/chat frontend	AMD, NVIDIA, Apple

Data & ML

Service	Description	GPU
`label-studio/`	Label Studio — data labeling for ML training	CPU
`anythingllm/`	AnythingLLM — all-in-one RAG + chat + agents	AMD, NVIDIA

Platform Compatibility

Quick reference for hardware requirements. Data sourced from each service's manifest.yaml.

Service	NVIDIA	AMD	Apple Silicon	CPU Only	Min VRAM
aider	✓	✓	✓	—	—
anythingllm	✓	✓	—	—	2 GB
audiocraft	✓	—	—	—	6 GB
bark	✓	—	—	—	4 GB
baserow	—	—	—	✓	—
chromadb	✓	✓	✓	—	—
continue	✓	✓	✓	—	4 GB
crewai	—	—	—	✓	—
dify	✓	✓	—	—	4 GB
flowise	✓	✓	✓	—	—
fooocus	✓	—	—	—	8 GB
forge	✓	—	—	—	8 GB
frigate	✓	—	—	—	1 GB
gaia	—	—	—	✓	—
gitea	—	—	—	✓	—
immich	✓	✓	—	—	2 GB
invokeai	✓	✓	—	—	8 GB
jan	✓	✓	—	—	—
jupyter	✓	✓	—	—	4 GB
label-studio	—	—	—	✓	—
langflow	✓	✓	✓	—	—
librechat	✓	✓	✓	—	—
localai	✓	✓	—	—	4 GB
milvus	—	—	—	✓	—
ollama	✓	✓	—	—	8 GB
open-interpreter	—	—	—	✓	—
paperless-ngx	—	—	—	✓	—
piper-audio	✓	✓	✓	—	—
rvc	✓	✓	—	—	6 GB
sillytavern	✓	✓	✓	—	—
text-generation-webui	✓	✓	—	—	4 GB
weaviate	—	—	—	✓	—
xtts	✓	✓	—	—	—

Note: "Min VRAM" shows the minimum GPU memory needed for the service's base feature. Some advanced features may require more. Services marked "CPU Only" have no GPU acceleration. Apple Silicon support means the service can use Metal/MPS for acceleration.

Workflows

Pre-built automation workflows for n8n, ComfyUI, Flowise, Langflow, and more:

workflows/
├── bark/          — Voice synthesis pipelines
├── comfyui/       — LLM-to-image generation
├── flowise/       — Chatflow API templates
├── langflow/      — Flow API templates
├── n8n/           — Webhooks, scheduling, form processing, DB sync
├── piper/         — TTS conversion
├── rvc/           — Voice conversion
├── sillytavern/   — Status monitoring
├── text-generation-webui/  — Chat completion, model listing
└── whisper/       — Speech-to-text conversion

Templates

Everything you need to build your own extension:

Template	Purpose
`service-template.yaml`	Complete manifest with every field documented
`compose-template.yaml`	Compose fragment with best practices
`compose-gpu-only.yaml`	GPU-only service pattern (no CPU fallback)
`compose-gpu-swap.yaml`	CPU base + GPU overlay pattern
`dashboard-plugin-template.js`	Dashboard UI plugin scaffold

Schema

schema/service-manifest.v1.json — JSON Schema for validating manifest files.

# Validate any manifest:
python3 -c "
import json, yaml
from jsonschema import validate
schema = json.load(open('schema/service-manifest.v1.json'))
manifest = yaml.safe_load(open('services/bark/manifest.yaml'))
validate(manifest, schema)
print('Valid!')
"

How to Use

Add a service to your ODS:

# Copy the extension into your ODS
cp -r ods/extensions/library/services/bark ods/extensions/services/bark

# Enable and start it
ods enable bark
ods start bark

Build your own extension:

# Copy the template
cp -r ods/extensions/library/templates/ my-service/
mv my-service/service-template.yaml my-service/manifest.yaml
mv my-service/compose-template.yaml my-service/compose.yaml

# Edit, then validate
python3 -c "import yaml; yaml.safe_load(open('my-service/manifest.yaml'))"

Status

These extensions are actively tested on ODS development builds. Some are battle-tested (Ollama, ChromaDB, Bark), others are newer. All follow the v1 manifest schema and integrate with the ODS service registry, dashboard, and CLI.

17 services have already graduated to production — these 32 are being prepared for the next wave.