AI templates
Model runners, chat front ends and AI tooling. Some entries are marked as needing an NVIDIA GPU to be practical, which the page says plainly; the rest run on CPU.
- AgentgatewayProxy built for agent traffic that routes MCP, A2A and LLM requests. The install form carries config.yaml as an editable file, so it starts with a listener that answers on port 3000.
- AnythingLLMA private chat workspace over your documents, with built-in RAG and support for local or hosted models.
- BifrostFast open source LLM gateway that routes requests to many providers behind one OpenAI compatible API, with a web UI for keys and routing rules.
- Big-AGIA multi-model chat workspace with personas, side by side answers and voice, using your own API keys. MIT licensed. Keys stay in your browser unless you set them on the server.
- ComfyUIA node-graph editor for building and running Stable Diffusion and other image generation pipelines.
- Crawl4AIAn open source web crawler API that turns pages into clean Markdown and structured data for LLM pipelines, with a built-in playground and MCP endpoint. API calls need the generated token as a Bearer header.
- Docling ServeAn API server that converts PDFs, Office files and images into Markdown, JSON or text for RAG pipelines, using the Docling library on CPU. MIT licensed. The image is about 5 GB, the interactive UI is at /ui and the API docs at /docs. There is no login, so keep it private or behind an auth proxy.
- FlowiseDrag-and-drop builder for LLM chatbots, agents, and RAG flows, with an API for each flow.
- Flowise (Postgres + Qdrant)Flowise with a Postgres record manager, a Redis cache, and a Qdrant vector store, for production RAG flows.
- HollamaA minimal web chat client for Ollama and OpenAI compatible servers. MIT licensed. Add the address of your model server in settings.
- Jupyter GPU NotebookJupyterLab with a CUDA-ready data science stack for training and experimenting with models.
- Kokoro TTSA CPU text-to-speech server with an OpenAI-compatible /v1/audio/speech API, built on the Kokoro model. Apache-2.0 licensed. The image is about 3 GB and there is no login, so keep it on a private network or behind an auth proxy.
- Label StudioA flexible data labeling tool for text, images, audio, and video, with export to common ML formats.
- LangflowA visual low-code tool for building and deploying multi-agent and RAG workflows in Python.
- LangfuseTrace, evaluate, and debug LLM applications with prompt management and cost analytics.
- LibreChatA self-hosted ChatGPT-style interface for many model providers, with agents, search, and multi-user login.
- LightRAGGraph based retrieval augmented generation server with a web UI and API. Runs on local file storage; point it at a model provider with the LLM and embedding settings.
- LiteLLM ProxyOne OpenAI-format gateway in front of 100+ model providers, with keys, budgets, and spend tracking.
- llama.cpp ServerThe llama.cpp HTTP server: run quantized GGUF models on plain CPUs with an OpenAI-compatible API.
- Lobe ChatA modern, open-source AI chat framework supporting many model providers.
- LocalAIA drop-in OpenAI-compatible API that runs language, image, and speech models on your own hardware.
- MetaMCPAn MCP aggregator and gateway that groups many MCP servers behind one endpoint with auth, namespaces and middleware.
- MindsDBA query layer that connects data sources and AI models so you can ask questions in SQL. MIT licensed. Includes the web editor and the HTTP API.
- MLflowTrack experiments, compare runs, and manage models with a self-hosted MLflow tracking server.
- n8n AI Starter Kitn8n wired to a local Ollama model and a Qdrant vector store, ready for private AI workflows.
- n8n MCPA Model Context Protocol server that gives AI assistants searchable n8n node documentation. MIT licensed. The bearer token is generated for you.
- New APIAn LLM gateway and AI asset management system that unifies many model providers.
- OllamaA local LLM runtime with an HTTP API. Pull any model yourself once it's running.
- Ollama + Open WebUIA local LLM runtime paired with a ChatGPT-style web interface, for running open models on your own hardware.
- Ollama: DeepSeek-R1 7BDeepSeek's R1 distilled 7B reasoning model, served locally through Ollama's HTTP API.
- Ollama: Llama 3 8BMeta's Llama 3 8B model, served locally through Ollama's HTTP API.
- Ollama: Mistral 7BMistral's 7B instruct model, served locally through Ollama's HTTP API.
- Ollama: Phi-3 MiniMicrosoft's Phi-3 Mini (3.8B), the smallest model in this catalog, good for constrained nodes.
- Ollama: Qwen 2.5 7BAlibaba's Qwen 2.5 7B model, served locally through Ollama's HTTP API.
- Open NotebookAn open source, self-hosted alternative to NotebookLM for researching with your own sources, with podcasts and chat. Add model provider keys in the app after install.
- Open WebUIA self-hosted chat interface for running local large language models through Ollama.
- OpenClawA self-hosted personal AI agent gateway that connects chat apps and tools to the model provider you choose. Recommended container limit is 2 GiB. Measured use is 0.9 to 1.3 GiB, so a Levelrail VM needs at least 4096 MiB. The gateway trusts the ingress proxy through gateway.trustedProxies, written once to openclaw.json in the data volume.
- Paperless-AIAdds automatic tagging, titles and document chat to a Paperless-ngx server using your choice of AI provider. MIT licensed. A setup wizard collects the Paperless address and token.
- PromptfooTest and compare prompts, models and agents with a local web UI for browsing evaluation results. Runs the viewer with results stored on a volume.
- SillyTavernA local chat front end for language models with characters, lorem books and extensions, bringing your own model API. AGPL-3.0 licensed. Sign in with the admin user and the generated password.
- SpeachesAn OpenAI-compatible speech server: Whisper transcription and text-to-speech behind one API.
- Stable Diffusion WebUIA browser interface for generating and editing images with Stable Diffusion checkpoints.
- Text Embeddings InferenceA lean server that turns text into embeddings with popular open models, on CPU with no extra setup.
- Text Generation InferenceHugging Face's production inference server for transformer language models, with continuous batching and streaming.
- UnstructuredAn API that turns PDFs, Word files, and other documents into clean data for RAG pipelines.
- vLLMA fast, high-throughput inference server that exposes an OpenAI-compatible API for open-weight language models.
- Wyoming openWakeWordLocal wake word detection over the Wyoming protocol for Home Assistant voice assistants, preloaded with the ok nabu model.
- Wyoming PiperFast local text to speech over the Wyoming protocol, used by Home Assistant voice assistants. Downloads the chosen voice on first start.