ansible-proxmox-ai

This repository provides a set of Ansible roles for building a homelab AI/LLM infrastructure on Proxmox VMs and LXC containers. It was extracted from a larger ansible-proxmox-apps collection and focuses specifically on the AI stack. The roles are organised into several functional areas: - **LLM serving**: ollama (model server), llama_cpp with llama-swap (GPU-tier), llm_router (LiteLLM proxy as the OpenAI-compatible front door), and open_webui (chat frontend). - **RAG**: llamaindex (Python + Ollama CPU-only embeddings pipeline) and qdrant_docker (vector database). - **Agents**: hermes_agent (NousResearch agent gateway), agent_exec (sandboxed execution), agentgateway_docker, codex_runner (isolated Codex CLI user). - **LLM app platforms**: Dify, LangFlow, LangGraph, Langfuse, and Arize Phoenix (all Docker-based). - **Ops**: fabric_watchdog, a systemd timer that probes the agent fabric from the Hermes guest and alerts over Slack. Development tooling is managed via a Nix flake and direnv, providing ansible-core, ansible-lint, molecule, sops, and related utilities. Linting and molecule‑based testing (with Docker) are available for individual roles. Currently only roles are shipped; a site-level playbook and inventory loader are planned follow‑ups. The project uses conventional commits and a git‑flow branching model (develop branch). Licensed under Apache 2.0.