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 agentmods add skills/rhtevan/agentfs/hosted-model-ctlnpx skills add rhtevan/agentfs --skill hosted-model-ctlgit clone --depth 1 https://github.com/rhtevan/agentfsWrote 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/rhtevan/agentfs/hosted-model-ctl)<a href="https://agentmods.dev/skills/rhtevan/agentfs/hosted-model-ctl"><img src="https://agentmods.dev/badge/skills/rhtevan/agentfs/hosted-model-ctl.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 | $0.00048 | $0.01872 |
| Opus 5 | $0.00024 | $0.00936 |
| Sonnet 5 | $0.00010 | $0.00374 |
| Haiku 4.5 | $0.00005 | $0.00187 |
Grade A, and why
hosted-model-ctl 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 3d 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 — 184 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Hosted Model Control
Deploy, manage, and test self-hosted LLM model containers on NVIDIA GPUs via Podman. All deployments use profiles — curated, VRAM-validated configurations. No individual model aliases.
Hardware
| Host | SSH | GPUs | VRAM | Bandwidth | Port |
|---|---|---|---|---|---|
| rhtevan-work | rhtevan-work |
1× RTX A500 | 4 GB | 128 GB/s | 10000 |
| rhel-ai | rhel-ai |
4× NVIDIA L4 | 88 GB | 1,200 GB/s | 9000 |
Deployment Profiles
One profile active per host at a time (mutual exclusion). All profiles on the same host share a port — deploying a new profile automatically stops the current one.
Full details in references/deployment-profiles.md.
rhtevan-work (port 10000)
| Profile | Model | Engine | Context | Speed | Default |
|---|---|---|---|---|---|
g3b-16k |
Granite 3B Q4_K_M | llama.cpp | 16K | 37 tok/s | ✅ |
g350m-2k |
Granite 4.0 350M FP16 | vLLM | 2K | 44 tok/s |
rhel-ai (port 9000)
| Profile | Model | Engine | TP | Context | Speed | Default |
|---|---|---|---|---|---|---|
g8b-fp8-spec-128k |
Granite 8B FP8 + 3B FP8 draft | vLLM spec | 4 | 128K | 58-79 tok/s | ✅ |
g8b-spec-128k |
Granite 8B BF16 + 3B BF16 draft | vLLM spec | 4 | 128K | 19-25 tok/s |
Speculative Decoding
The g8b-spec-* profiles use draft model speculation:
a Granite 3B model generates candidate tokens, then the 8B target
verifies them in a single forward pass — converting wasted GPU compute
into useful token verification (2-6× speedup).
- g8b-fp8-spec-128k: FP8 weights + CUDA graphs (recommended, ~70 tok/s)
- g8b-spec-128k: BF16 weights +
--enforce-eager(safe fallback, ~20 tok/s)
Specification
| ID | Capability | Verifiable By |
|---|---|---|
| S1 | List all profiles with container status | scripts/list.sh → table output |
| S2 | Pre-check host readiness | scripts/pre-check.sh HOST → pass/fail report |
| S3 | Deploy profile (idempotent) | scripts/setup.sh PROFILE → container created, API HTTP 200 |
| S4 | Start existing profile (mutual exclusion) | scripts/start.sh PROFILE → running |
| S5a | Stop profile | scripts/stop.sh PROFILE → stopped, state cleared |
| S5b | Stop all containers across both hosts | scripts/stop.sh all → all exited |
| S6 | Show status (includes active profile) | scripts/status.sh [PROFILE] → status report |
| S7 | Test running model | scripts/test.sh PROFILE → 4 tests pass |
| S8 | Generate platform report | scripts/report.sh [HOST] → markdown report |
What ships with it
14 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- CHANGELOG.md 4.8 KB
- references/benchmark-report.md 19 KB
- references/deployment-profiles.md 4.6 KB
- references/memory-budget.md 4.8 KB
- references/model-landscape.md 10 KB
- scripts/common.sh 6.5 KB runs code
- scripts/list.sh 1.8 KB runs code
- scripts/pre-check.sh 2.7 KB runs code
- scripts/report.sh 17 KB runs code
- scripts/setup.sh 4.9 KB runs code
- scripts/start.sh 2.1 KB runs code
- scripts/status.sh 1.9 KB runs code
- scripts/stop.sh 2.5 KB runs code
- scripts/test.sh 2.7 KB runs code
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.
- 3d ago First seen · 184 lines · 48 tokens per session scan A 661d4b111284
hosted-model-ctl is a skill published in the GitHub repository rhtevan/agentfs (2 stars, last pushed 4d ago), licensed Apache-2.0. It adds 48 tokens to every session and 1,872 once invoked, about $0.0002 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
harbor
CLI toolkit for managing containerized LLM services. Use when the user wants to start, stop, configure, or manage AI/LLM services like Ollama, Open WebUI, llama.cpp, vLLM, LiteLLM, ComfyUI, and 250+ others. Triggers on requests to "run a model", "start ollama", "set up an LLM", "configure harbor", "manage services"…
airunway-aks-setup
Set up AI Runway on AKS — from bare cluster to running model. Covers cluster verification, controller install, GPU assessment, provider setup, and first deployment. WHEN: "setup AI Runway", "onboard AKS cluster", "install AI Runway", "airunway setup", "deploy model to AKS", "GPU inference on AKS", "KAITO setup on…
rag-blueprint
NVIDIA RAG Blueprint — deploy, configure, troubleshoot, and manage. Handles any RAG action: deploy, install, start, enable, disable, toggle, change, configure, troubleshoot, debug, fix, shutdown, stop, or tear down any RAG feature or service (Agentic RAG, VLM, guardrails, query rewriting, models, search, ingestion…
huggingface-spaces
Build, deploy, debug, or maintain a Hugging Face Space using Gradio, Docker, or Static SDKs. Use for general Space hosting and configuration; use huggingface-zerogpu for ZeroGPU runtime constraints and lora-space-builder for LoRA demos.
tao-run-automl
Run container-backed AutoML / hyperparameter optimization (HPO) for NVIDIA TAO networks using AutoMLRunner. Handles algorithm selection (bayesian, hyperband, asha, bohb, llm, hybrid, autoresearch), WandB experiment tracking, job execution on any TAO SDK platform, result interpretation, and per-rec custom evaluation…
nvwb
Manage NVIDIA AI Workbench projects, contexts, builds, and environments via the nvwb CLI. This skill should be used when the user wants to create, clone, build, manage, or configure AI Workbench projects, contexts, or environments.