Self-host AGmind
v3.2.0Private RAG platform installer for NVIDIA DGX Spark
- Stars
- 48
- CPU
- 20-core ARM (10× Cortex-X925 + 10× Cortex-A725)
- RAM
- 128 GB LPDDR5X unified memory
- Storage
- 100 GB+ free on /
- Ports
- 80 (nginx HTTP reverse proxy), 3001 (Grafana dashboards), 9443 (Portainer container management UI), 8000 (vLLM LLM API on peer node), 5678 (n8n workflow automation (optional, localhost only))
What is AGmind?
AGMind is a self-hosted installer for a private Retrieval-Augmented Generation platform built specifically for NVIDIA DGX Spark (GB10) systems with 128 GB of unified memory. It provisions over 30 containers through Docker Compose, including Dify for workflow orchestration, vLLM for large language model inference using gemma-4-26B, Weaviate or Qdrant for vector storage, RAGFlow for advanced document parsing, Docling for document extraction with OCR, and a complete monitoring suite with Prometheus, Grafana, Loki, and Alertmanager. The installer provides an interactive wizard with hardware autodetection, optional dual-Spark clustering over 200G QSFP, production hardening with UFW, fail2ban, Authelia for two-factor authentication, and capability dropping, plus a day-2 command-line interface for ongoing operations. The software runs on DGX OS 7.5.0 (Ubuntu 24.04 LTS arm64) with the NVIDIA 580.x driver pinned, and is licensed under Apache 2.0.
Should you self-host AGmind?
AGmind scores 41/100 on selfhost.directory (how scoring works) and ranks #97 of 125 in AI. It has 48 GitHub stars.
It shipped 3 stable releases in the past year. The latest, v3.2.0, came out on 20 May 2026. New versions land roughly every 2 weeks.
Setup is advanced: plan for several services and some ongoing maintenance. It is Other-licensed and written mostly in Shell.
Its closest alternative on selfhost.directory is ComfyUI (93/100, 143 668 stars). It scores higher overall, so compare the two before you pick. See AGmind vs ComfyUI side by side.
Features
- Bilingual wizard
- mDNS service discovery
- Non-interactive install
- Versioned state store
AGmind README
More in RAG
bishengAn Apache-2.0, self-hosted LLM devops platform for enterprise AI: GenAI workflows, RAG knowledge bases, agents, model management, evaluation, SFT and OCR.
CasibaseOpen-source personal AI assistant with RAG knowledge bases, MCP tool use and autonomous agent loops; one self-hosted binary with 30+ model providers, SSO and audit logs.
chromaOpen-source vector database for AI retrieval: dense, sparse, full-text, regex and metadata search. Self-host it with pip, the chroma CLI, or Docker on port 8000.
ConnapseConnapse is a self-hosted, open-source knowledge backend giving AI agents persistent, searchable memory via hybrid search, MCP server, and storage connectors.
DocBrainDocBrain is a self-hosted RAG framework with document Q&A, natural language SQL over tables, local vectors, multi-LLM support, and RBAC.
Docling StudioSelf-hosted web app for visual PDF analysis — configure extraction, inspect bounding boxes, chunk content, and index into OpenSearch or Neo4j.
fastgptFastGPT is a self-hosted, open-source AI agent platform for building RAG chat apps with visual workflow orchestration, hybrid retrieval and reranking.
fortemiHost your own AI knowledge base with hybrid search, knowledge graphs, multiple LLM backends, and multimodal ingestion. Docker deployment on regular hardware.
GaiaNet NodeOpen-source, self-hosted GaiaNet node: pick any Hugging Face LLM, add a RAG knowledge base, serve an OpenAI-compatible API and web chat UI.
graphitiOpen-source Python framework building temporal knowledge graphs for AI agents: hybrid semantic, keyword and graph search, facts with validity windows, episode provenance.
KhojSelf-hosted, open-source AI second brain: RAG chat over your notes and the web, custom agents, semantic search and scheduled research emailed to you.
kotaemonSelf-hosted open-source RAG chat UI for QA over your own documents, with hybrid retrieval, PDF citations, multi-user collections and local LLMs via Ollama.
0 comments