Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add chipi/agentic-ai-homelab --skill gpu-modegit clone --depth 1 https://github.com/chipi/agentic-ai-homelabWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/chipi/agentic-ai-homelab/gpu-mode)<a href="https://agentmods.dev/skills/chipi/agentic-ai-homelab/gpu-mode"><img src="https://agentmods.dev/badge/skills/chipi/agentic-ai-homelab/gpu-mode.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00117 | $0.00763 |
| Opus 5 | $0.00059 | $0.00381 |
| Sonnet 5 | $0.00023 | $0.00153 |
| Haiku 4.5 | $0.00012 | $0.00076 |
Grade A, and why
gpu-mode scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 61 lines — stays where its author put it; the contents beside it link to each section on GitHub.
gpu-mode
The DGX has one GB10 GPU; the coder-next and autoresearch vLLMs can't both
hold it (both want ~90% of VRAM). gpu-mode-swap.sh is the authoritative
coordinator. This skill wraps it: check freely, switch only on explicit go.
Connect
The script lives on the DGX host, symlinked at ~/bin/gpu-mode-swap.sh. Reach it
over the tailnet with ssh (non-interactive → always the absolute path; the
gpu-mode zsh alias is not loaded in agent shells):
ssh dgx-llm-1 '~/bin/gpu-mode-swap.sh <args>'
Check the mode — read-only, always safe
Before ANY local vLLM call, confirm the right stack owns the GPU:
ssh dgx-llm-1 '~/bin/gpu-mode-swap.sh --mode-only' # → code | research | idle | BROKEN-BOTH
ssh dgx-llm-1 '~/bin/gpu-mode-swap.sh status --json' # full machine-readable state
- code — coder-next up on
:9000 - research — autoresearch up on
:8003 - idle — both down; GPU free for ML training / Ollama
- BROKEN-BOTH — both listening (bad state); needs a human, do not pile on
If the mode already matches what you need, proceed. If not, see below — but do not switch on your own.
Switch the mode — SHARED-STATE, explicit approval required every time
Switching brings one vLLM up and the other down; it interrupts whatever is using the GPU right now (a sweep, a training job, Ollama). Treat the DGX as production:
- Never switch as a side effect of "I need the endpoint." Ask first.
idledoes not mean the GPU is idle — a training job or Ollama can hold it directly while gpu-mode-swap reportsidle. Confirm with the operator before taking the GPU.- Only after an explicit go, and by absolute path:
ssh dgx-llm-1 '~/bin/gpu-mode-swap.sh code --json' # coder-next up, autoresearch down ssh dgx-llm-1 '~/bin/gpu-mode-swap.sh research --json' # autoresearch up, coder-next down ssh dgx-llm-1 '~/bin/gpu-mode-swap.sh idle --json' # both down
Report
- State the current mode plainly. If a switch happened, show the
--jsonresult and the resulting mode. - On
BROKEN-BOTHor a failedwait_for_port, stop and surface it — follow the troubleshooting indocs/recipes/gpu-mode-swap.md; do not retry blindly.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago First seen · 61 lines · 117 tokens per session scan A e8212f849259
gpu-mode is a skill published in the GitHub repository chipi/agentic-ai-homelab (2 stars, last pushed 4d ago), licensed MIT. It adds 117 tokens to every session and 763 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
agent-v3-memory-specialist
Agent skill for v3-memory-specialist - invoke with $agent-v3-memory-specialist.
agent-neural-network
Agent skill for neural-network - invoke with $agent-neural-network.
embeddings
Vector embeddings with HNSW indexing, sql.js persistence, and hyperbolic support. 75x faster with agentic-flow integration. Use when: semantic search, pattern matching, similarity queries, knowledge retrieval. Skip when: exact text matching, simple lookups, no semantic understanding needed.
neural-training
Neural pattern training with SONA (Self-Optimizing Neural Architecture), MoE (Mixture of Experts), and EWC++ for knowledge consolidation. Use when: pattern learning, model optimization, knowledge transfer, adaptive routing. Skip when: simple tasks, no learning required, one-off operations.
spark-training-gotchas
Preflight and diagnose the ten known failure modes for ML training on NVIDIA DGX Spark. Use when a training run on DGX Spark fails to start, OOMs below the 128GB limit, slows down mid-run, or before any multi-hour training job on GB10.
cupynumeric-migration-readiness
Pre-migration readiness assessor for porting NumPy to cuPyNumeric. Use BEFORE substantial porting work begins when the user asks whether code will scale on GPU, whether they should migrate to cuPyNumeric, which NumPy patterns transfer cleanly, what must be refactored before porting, or mentions pre-port assessment…