AWS Startups is an official AWS repository containing plugins, skills, tools, and other resources for people building startup products on Amazon Web Services. Its add-ons support startup-focused architecture, migration, and development work on AWS.
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 awslabs/startups --skill self-hosted-llm-batch-inferencegit clone --depth 1 https://github.com/awslabs/startupsWrote 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/awslabs/startups/self-hosted-llm-batch-inference)<a href="https://agentmods.dev/skills/awslabs/startups/self-hosted-llm-batch-inference"><img src="https://agentmods.dev/badge/skills/awslabs/startups/self-hosted-llm-batch-inference/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/awslabs/startups/self-hosted-llm-batch-inference"><img src="https://agentmods.dev/badge/skills/awslabs/startups/self-hosted-llm-batch-inference.svg" alt="Reviewed on agentmods" width="80" 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 | $0.00170 | $0.01159 |
| Opus 5.5 | $0.00068 | $0.00464 |
| Sonnet 5.5 | $0.00034 | $0.00232 |
| Haiku 4.5 | $0.00017 | $0.00116 |
Grade A, and why
self-hosted-llm-batch-inference 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 1mo 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 — 88 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Self-Hosted LLM Batch Inference
Build a repeatable offline inference path for a startup that needs GPU economics without operating an ML platform. Use one AWS Batch job for one corpus run, backed by Amazon ECS Managed Instances Spot. Stage model weights in Amazon S3, divide the corpus into deterministic shards, and process those shards sequentially with vLLM on one GPU.
Keep the infrastructure stack separate from each run. Provision reusable Batch, ECR, IAM, logging, and optional networking resources with infrastructure as code. Submit each run separately with immutable image, model, prompt, input manifest, and output identifiers.
Set scaleInAfter to 0 for this architecture. No follow-up task is expected on the instance after
the Batch job ends, so retaining idle GPU capacity adds cost without improving the run.
Scope
Apply this skill only when all of these conditions hold:
- The workload is recurring, latency-tolerant, generative offline inference.
- The model is approximately 1B to 8B parameters and fits on one GPU after measurement.
- The team accepts Spot interruption and requires automatic recovery.
- One model is loaded once and reused across many input shards in a corpus run.
- Persistent artifacts and completion records can live in Amazon S3.
Do not apply this skill to online endpoints, training or fine-tuning, embedding models, other inference engines, multi-GPU or distributed inference, or one-off experiments where an operator can rerun the command manually.
Workflow
- Read
references/prepare-model-and-input.mdand define the immutable model artifact, prompt contract, deterministic shard manifest, output schema, quarantine schema, and measured sizing inputs. - Read
references/run-with-aws-batch.mdand author the reusable infrastructure plus the separateSubmitJobinvocation. Preserve one Batch job per corpus run. - Read
references/verify-and-optimize.mdbefore the canary and again after the full run. Reconcile records, confirm retry behavior, measure cost inputs, and verify that GPU capacity has terminated.
What ships with it
3 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.
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.
- 1mo ago First seen · 88 lines · 170 tokens per session scan A 5bc070c185d8
self-hosted-llm-batch-inference is a skill published in the GitHub repository awslabs/startups (24 stars, last pushed 12d ago), licensed Apache-2.0. It adds 170 tokens to every session and 1,159 once invoked, about $0.0007 per session on Opus 5.5. 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 2026-09-12.
Other skills, from other repositories
tensorrt-llm
High-throughput LLM inference on NVIDIA GPUs.
agent-platform-tuning
Agent Platform Model Tuning. Use when you need to fine-tune open models or Gemini models using Agent Platform infrastructure. Don't use for model training outside Agent Platform, model deployment to endpoints (use agent-platform-deploy), or managing serving endpoints (use agent-platform-endpoint-management).
google-cloud-solution-guided-gke-ai-migration
Guides the migration of existing AI workloads (Cloud Run, Gemini API, Gemini Enterprise Agent Platform) to self-hosted GKE inference using gcloud and kubectl. Use when the user has an existing AI inference workload (on Cloud Run, the Gemini API, Gemini Enterprise Agent Platform, or a custom VM) and wants to move it to…
modal
Modal is a serverless cloud platform for running Python on demand, including on-demand GPUs. Use when deploying or serving AI/ML models, running GPU-accelerated workloads (training, fine-tuning, inference), serving web endpoints, scheduling batch jobs, or scaling Python code to cloud containers with the Modal SDK.
agent-platform-endpoint-management
Manages Agent Platform serving endpoints. Use when you need to create, list, describe, update, or delete serving endpoints for model deployment on Agent Platform. Also use when troubleshooting endpoint permission, quota, or resource busy errors. Don't use for deploying models to endpoints or for running model…
gke-inference
Deploys and optimizes AI/ML inference workloads on GKE, using GPUs, TPUs, and model servers. Use when deploying GKE inference servers, configuring GKE GPU resources for inference, or deploying LLMs on GKE. Don't use for migrating existing AI workloads to GKE (use google-cloud-solution-guided-gke-ai-migration), GKE RAG…