{"name":"compute.md API","version":"1.0","description":"Read-only compute data for agents. No key, CORS open. compute.md never books, bills or holds credentials.","disclaimer":"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.","terms":"https://compute.md/terms","skill":"https://compute.md/compute.md","openapi":"https://compute.md/openapi.json","mcp":{"url":"https://compute.md/mcp","transport":"Streamable HTTP (stateless, JSON responses, no auth)","tools":["fit_model","gpu_specs","model_spec"],"prompts":["compute (optional lever arguments)"]},"endpoints":[{"path":"/compute.md","summary":"The COMPUTE.md skill. Add any lever to get a version with a Your limits section written in. Every lever is a number or a fixed choice.","format":"markdown","params":[{"name":"budget_usd","type":"number","description":"Budget per task ($), 1 to 1000000."},{"name":"approval_usd","type":"number","description":"Ask before any spend over ($), 0 to 1000000."},{"name":"max_hourly_usd","type":"number","description":"Most per hour, all machines ($), 0.1 to 100000."},{"name":"model_budget_usd","type":"number","description":"Model calls per task ($), 0 to 1000000."},{"name":"stop_hours","type":"number","description":"Hard stop after (hours), 0.25 to 720."},{"name":"idle_minutes","type":"number","description":"Shut down when idle for (minutes), 5 to 1440."},{"name":"checkpoint_minutes","type":"number","description":"Checkpoint every (minutes), 5 to 240."},{"name":"spot","type":"string","enum":["never","checkpointed","preferred"],"description":"Spot or interruptible"},{"name":"providers","type":"string","enum":["existing","ask"],"description":"Providers"},{"name":"regions","type":"string","enum":["any","data","us","eu"],"description":"Regions"},{"name":"approver","type":"string","enum":["user","owner","requester"],"description":"Who approves"}],"examples":["https://compute.md/compute.md","https://compute.md/compute.md?budget_usd=100&approval_usd=25&stop_hours=6&spot=never"]},{"path":"/api/v1/fit","summary":"Size a model or job: memory, which GPUs fit and how many, and estimated speed. Estimates only; no prices.","format":"json","params":[{"name":"model","type":"string","description":"Hugging Face repo id (Qwen/Qwen3.6-27B) or a quick pick (gpt-oss-120b). Or pass params_b instead."},{"name":"task","type":"string","enum":["serve","lora","qlora","full"],"description":"Default serve."},{"name":"precision","type":"string","enum":["native","bf16","fp8","int8","int4"],"description":"Weight precision. Default native: the checkpoint as published."},{"name":"kv","type":"string","enum":["bf16","fp8"],"description":"KV cache precision. Default bf16."},{"name":"context","type":"string","description":"Tokens per sequence, such as 32768 or 32k. Default 32k serving, 4k training."},{"name":"concurrency","type":"integer","description":"Sequences in flight when serving, batch size when training. Default 8 or 4."},{"name":"train_tokens","type":"string","description":"Total training tokens (dataset x epochs), such as 50m, for an estimated duration."},{"name":"max_gpus","type":"integer","description":"Largest device count to consider. Default 8."},{"name":"gpu","type":"string","description":"Only these accelerators, comma-separated: h100-sxm,h200,mi300x."},{"name":"limit","type":"integer","description":"Options to return. Default 12."},{"name":"params_b","type":"number","description":"Parameters in billions, for a model that isn't on Hugging Face. Add active_b, layers, kv_heads and head_dim for speed and KV cache."}],"examples":["https://compute.md/api/v1/fit?model=Qwen/Qwen3.6-27B&context=32k&concurrency=8","https://compute.md/api/v1/fit?model=meta-llama/Llama-3.3-70B-Instruct&precision=fp8&kv=fp8&concurrency=32","https://compute.md/api/v1/fit?model=Qwen/Qwen3.5-9B&task=lora&train_tokens=50m"]},{"path":"/api/v1/gpus","summary":"Accelerator specs from vendor datasheets: memory, bandwidth, dense BF16 and FP8 throughput, interconnect. No prices.","format":"json","params":[{"name":"gpu","type":"string","description":"Accelerator ids or names, comma-separated: h100, h200, b200, a100, l40s, mi300x, rtx-4090, tpu-v6e."},{"name":"min_memory_gb","type":"number","description":"Only accelerators with at least this much memory."},{"name":"vendor","type":"string","description":"Only one vendor: NVIDIA, AMD, Google, AWS, Intel, Apple."}],"examples":["https://compute.md/api/v1/gpus","https://compute.md/api/v1/gpus?min_memory_gb=140"]},{"path":"/api/v1/models","summary":"Model specs: parameters, active parameters, checkpoint size, attention layout and KV cache per token. Quick picks, or any Hugging Face repo via id.","format":"json","params":[{"name":"id","type":"string","description":"Hugging Face repo id. Omit for the quick-pick list."}],"examples":["https://compute.md/api/v1/models","https://compute.md/api/v1/models?id=openai/gpt-oss-120b"]}],"sources":[{"data":"Model sizes and architectures","name":"Public Hugging Face model repositories (model info, file sizes, config.json)","url":"https://huggingface.co","refresh":"read on request, cached for a day"},{"data":"Accelerator specs","name":"Vendor datasheets, collected by hand","url":"https://compute.md/api/v1/gpus","refresh":"by hand"}],"usage":"Calls are counted anonymously (no IPs or personal data stored) to see what is useful."}