Skip to content

Releases: botAGI/AGmind

AGMind v3.0.2 — DGX Spark RAG platform

28 Apr 19:40
@botAGI botAGI
v3.0.2
37cc0cd

Choose a tag to compare

View all tags
AGMind v3.0.2 — DGX Spark RAG platform Latest
Latest

First public release of AGMind — a one-command installer for a private RAG platform on NVIDIA DGX Spark (GB10 / aarch64).

What's inside

Dify + vLLM + Weaviate/Qdrant + RAGFlow + Docling-serve cu130 + monitoring (Prometheus + Grafana + Loki + Alloy + Portainer 2.39.1).

Highlights

  • DGX Spark only — x86_64 path retired, NVIDIA driver 580 pinned (UMA leak / CUDAGraph deadlock on 590+/595+).
  • RAGFlow v0.24.1-spark — deep document parsing + GPU OCR (Latin/Cyrillic/Chinese), pre-built ar2r223/ragflow-spark image; 7 ingestion templates translated to RU.
  • Dual-Spark cluster — master + peer over QSFP 200G DAC, NAT-on-demand for peer WAN, Portainer peer agent auto-deployed, peer 4 TB SSD as remote backup target.
  • Monitoring — Promtail → Grafana Alloy migration, 10 Grafana dashboards, custom textfile collector for agmind_gpu_* (NVML returns N/A on GB10 unified memory).
  • GB10 unified memory aware — calibrated mem_limit / gpu_memory_utilization for the 121 GiB shared pool; FlashInfer FP8 broken on SM_121, defaults to VLLM_ATTENTION_BACKEND=TRITON_ATTN.
  • Day-2 CLI — agmind status / health / backup / update / ragflow / docling bench / plugin-daemon / mdns-status; agmind upgrade --diff (pinned vs live).
  • Reliability — model caches preserved across re-install, SurrealDB password persistence, Redis stale-state guards on container recreate.

Install

git clone https://github.com/botAGI/AGmind.git
cd AGmind && sudo bash install.sh

See README.md for hardware requirements + driver pinning notes.

🤖 Auto-generated by gh release create — version metadata syncs to templates/release-manifest.json via the sync-release.yml workflow.

Assets 3
Loading