Self-hosted LLM/RAG stack in one command — AMD Strix Halo / x86_64 (ROCm/Vulkan, Docker Compose)

Reproducible local-LLM benchmark harness: llama.cpp on AMD Strix Halo (gfx1151, Ryzen AI Max+ 395) and NVIDIA DGX Spark — frozen corpora, quality gates with unit tests, sealed run bundles. Apache-2.0

Open evidence archive for local-LLM measurements: 75 sealed run bundles from AMD Strix Halo (gfx1151) and DGX Spark — per-request records, manifests, checksums, and the claim catalog behind agmind.…

MCP server for measured local-LLM benchmarks from the AGmind claim registry — AMD Strix Halo (Ryzen AI Max+ 395) llama.cpp numbers today, NVIDIA DGX Spark on the bench. search_claims, get_claim, li…

Measured LLM benchmarks for NVIDIA DGX Spark (GB10): DeepSeek-V4-Flash 284B MoE on a TP=2 pair over 200G RoCE — tok/s by profile and concurrency, 1M-token context curve, the MoE backend flag, monit…

Measured LLM benchmarks for AMD Strix Halo / Ryzen AI Max+ 395 (Radeon 8060S, 128 GB unified): llama.cpp Vulkan & ROCm — decode pace, TTFA, prompt cache, quants, sustained load. Every number links …

Point it at a local inference server: reports what it can actually do — measured, not claimed. Catches capacity a server silently fails to deliver.

Proof-carrying cyber immunity for Linux containers: signed evidence, deterministic policy, local approval and exact TTL containment.

DeepSeek-V4-Flash-0731 on 2x NVIDIA DGX Spark (GB10): serving recipe, benchmark harness, measured results. The MoE backend flag that auto-selection misses.

Locally fine-tuned Russian RAG models: document splitter, query expansion, retrieval embedder. Teacher distillation → GGUF → llama.cpp on AMD Vulkan. Commercial-OK licenses.

Legacy: self-hosted LLM/RAG stack installer for DGX Spark (arm64). The work continued as AGmind Systems Lab — measured local-LLM performance evidence (Strix Halo, DGX Spark, llama.cpp, vLLM): https…

WAKE.md for AI agents: compile project state so agents stop starting cold.

Multi-slot LLM inference on AMD Strix Halo: recipes + honest benchmarks (236 tok/s @ 32 streams, llama.cpp Vulkan)

DeepSeek-V4-Flash DSpark speculative decoding on 2x DGX Spark (GB10/sm_121), 1M context — our fp8 recipe + independent reproduction and cross-build benchmarks of the NVFP4-KV build. Honest, apples-…

One-command local AI/RAG installer for macOS (Metal): Dify, Open WebUI, Ollama, Weaviate/Qdrant, Postgres, Redis. 230 tests.