# COMPUTE.md

> A skill that makes AI agents careful compute buyers: size the work before renting anything, get prices from your own providers, set up the stack you want, launch only inside rules you set, and tear everything down. Version 2.1. MIT.

The skill works offline in any agent. 126 settings in 15 groups are written into the file as two sections. Your stack says what to provision: workload and replicas, machine class and vCPUs, RAM (DDR4 or DDR5, ECC), GPUs (count, memory, class, interconnect, nodes), disks (NVMe, block, IOPS, throughput), object and file storage, databases, cache and queues, network (bandwidth, static IPs, IPv6, load balancer, TLS, CDN), OS and images, orchestration, models (task, context, latency, concurrency) and monitoring. Your rules say what the agent may do: budgets, stop rules, caps, exposure, remote access, model-call limits, encryption, secrets, permissions and regions. Every setting is a number or a fixed choice. Presets (core, recommended, locked, web_app, inference, gpu_job, data_pipeline, dev_box, static_site) give a starting point. An optional read-only API and MCP server write your stack and rules, check a plan against both before launch, and add sizing estimates and accelerator specs. compute.md lists no prices, its settings and plan checks name no providers (the sizing data names accelerators and open models), and it never books, bills, stores limits or plans, or holds credentials.

Use at your own risk. compute.md is provided as is, without warranty. Information here may be inaccurate, incomplete or not current, and is not verified; sizing figures are estimates. Not financial, legal or procurement advice. Confirm everything with your providers before you rent or buy.

## Skill

- [COMPUTE.md](https://compute.md/compute.md): load into your agent's instructions or save as ~/.claude/skills/compute/SKILL.md (sha256 b2f755936cfd6bf17bef069901bcbab5a8c0a6560bfaa86f189f21ea0027f15f)
- [COMPUTE.md with your stack and rules](https://compute.md/compute.md?preset=web_app&budget_usd=100&memory_gb=8&disk_type=nvme): a preset plus a few settings; each setting overrides the preset

## API (no key, CORS open)

- [Settings](https://compute.md/api/v1/levers): every setting with its role (stack or rule), unit, range, options and default, plus the presets
- [Stack and rules](https://compute.md/api/v1/limits?preset=web_app&budget_usd=200): build Your rules and Your stack as JSON, or Markdown with format=markdown
- [Check](https://compute.md/api/v1/check): POST {"limits": {...}, "plan": {...}} to check a provider-neutral plan against your stack and rules before launch; GET for an example
- [Fit](https://compute.md/api/v1/fit?model=Qwen/Qwen3.6-27B&context=32768&concurrency=8): memory, the GPUs that fit and estimated speed for serving or fine-tuning any Hugging Face model
- [GPU specs](https://compute.md/api/v1/gpus): memory, bandwidth and throughput from vendor datasheets
- [Models](https://compute.md/api/v1/models): quick-pick model specs; ?id=org/name for any Hugging Face repo
- [API index](https://compute.md/api/v1) and [OpenAPI](https://compute.md/openapi.json)

## MCP

- https://compute.md/mcp: Streamable HTTP, stateless, no auth. Tools: fit_model, gpu_specs, model_spec, build_limits, check_plan. Prompt: compute, with an optional preset and any setting. Resources: the skill and the settings catalog.

## Terms

- [Terms of use and disclaimer](https://compute.md/terms)

## Optional

- [Full text](https://compute.md/llms-full.txt): the skill plus the API reference and every lever
- [Manifest](https://compute.md/.well-known/compute.json)