the provider catalog

NVIDIA

{"meta":{"title":"NVIDIA NIM · hosted now, self-hosted later","angle":"The NIM catalog hosted: try models from many vendors on one endpoint, then self-host the same containers on your own GPUs when it gets serious.","pick":"evaluating across vendors with a self-hosting exit already on the map.","console":{"label":"build.nvidia.com","href":"https://build.nvidia.com"},"setup":[{"text":"Get an API key from any model page on the catalog."},{"text":"The key rides the environment, never the file","code":"export NVIDIA_API_KEY=…"}]},"usage":{"id":"nvidia","yaml":"nika: v1\nworkflow:\n  id: nim-catalog-call\n  description: \"one catalog, many vendors' models — the hosted endpoint and the container match\"\n\ntasks:\n  note:\n    infer:\n      model: nvidia/meta/llama-3.1-70b-instruct\n      max_tokens: 120\n      system: Answer like a release note, not an ad.\n      prompt: |\n        One sentence: what does self-hosting the same container\n        later buy a team that starts on a hosted endpoint?\n\noutputs:\n  line: ${{ tasks.note.output }}\n","file":"nvidia.nika.yaml"},"audit":{"clean":true,"models_resolve":true,"tasks":1,"waves":1,"cost_ceiling_usd":0,"cost_floor_usd":0,"pricing":[{"model":"nvidia/meta/llama-3.1-70b-instruct","input_per_million":0,"output_per_million":0}],"pricing_as_of":"2026-07-07"}}

[{"slug":"meta-llama-3-1-70b-instruct","id":"meta/llama-3.1-70b-instruct","aliases":["llama-70b"]},{"slug":"nvidia-nemotron-3-nano-30b-a3b","id":"nvidia/nemotron-3-nano-30b-a3b","aliases":["cheap"]},{"slug":"nvidia-nemotron-3-super-120b-a12b","id":"nvidia/nemotron-3-super-120b-a12b","aliases":["default"]}]

The NIM catalog hosted: try models from many vendors on one endpoint, then self-host the same containers on your own GPUs when it gets serious. One provider/model line moves a workflow here or away again: the same file runs on every provider the catalog names, and swapping is a one-word diff. Machines read the catalog; the binary answers nika catalog.

3pinned modelsaliases pin exact upstream ids
cloudfamilybring your own key; it rides an env var and stays yours
NVIDIA_API_KEYthe key rides herean env var, never a config file
openai-chatwire dialectthe adapter speaks it

the wire

cloud · bring your own key; it rides an env var and stays yours

A workflow names this provider in the model line (model: nvidia/nvidia/nemotron-3-super-120b-a12b), and the engine speaks openai-chat on the wire. The key rides NVIDIA_API_KEY in the environment, never a config file, never the workflow. The provider id also answers to nim · nvidia-nim.

default: nvidia/nemotron-3-super-120b-a12b
cheap: nvidia/nemotron-3-nano-30b-a3b
dialect: openai-chat
tags: function-calling · open-source · streaming

the models

3 pins

Aliases are what the file may write; the engine pins each one to an exact upstream id, so a run stays reproducible after a vendor rotates names.

defaultnvidia/nemotron-3-super-120b-a12bctx 262k · json:schema
cheapnvidia/nemotron-3-nano-30b-a3bctx 262k · json:schema
llama-70bmeta/llama-3.1-70b-instructctx 128k · json:schema$0/M in · $0/M out

prices are the ENGINE's, not ours: the audit's pricing snapshot (models.dev · pinned 2026-07-07): the same table nika check prices your ceiling with.

in the released catalog

3 rooms

The wire models this provider serves in the released binary's catalog · each one is a room with its seats, exact prices and measured energy.

meta/llama-3.1-70b-instructas llama-70b
nvidia/nemotron-3-nano-30b-a3bas cheap
nvidia/nemotron-3-super-120b-a12bas default

the whole register: /catalog/models · what the released binary knows, vendored at the engine pin.

in a real file

a complete workflow · shown whole

nvidia.nika.yaml

1nika: v12workflow:3  id: nim-catalog-call4  description: "one catalog, many vendors' models — the hosted endpoint and the container match"56tasks:7  note:8    ◇infer:9      model: nvidia/meta/llama-3.1-70b-instruct10      max_tokens: 12011      system: Answer like a release note, not an ad.12      prompt: |13        One sentence: what does self-hosting the same container14        later buy a team that starts on a hosted endpoint?1516outputs:17  line: ${{ tasks.note.output }}18

check-green against the released engine · 1 task · 1 wave · worst-case ceiling $0.0000 · the drift gate re-proves this copy on every test run · open it in the playground →

get set up

keys at build.nvidia.com

1Get an API key from any model page on the catalog.
2The key rides the environment, never the file export NVIDIA_API_KEY=…

Pick it for evaluating across vendors with a self-hosting exit already on the map. Then prove the seat: nika doctor names every provider this binary can reach.

swap it

10 more cloud seats

Moving off NVIDIA is editing one line. The rest of the file (tasks, permits, outputs, the trace) does not change. The register orders the catalog local first; beyond it the engine embeds 21 more OpenAI-compatible endpoints.

model: nvidia/nvidia/nemotron-3-super-120b-a12b model: ollama/qwen3.5:4bthe whole migration · nothing else moves

cross-references

the words that meet it

[]

where it lives

Try it in the playground, or install and go local first. Read the spec →