NVIDIA NIM: API key and base URL

NVIDIA NIM hosted inference. Bring your own key; nothing is billed by us.

What the harness already knows

Provider idnvidia
Bundled base URLhttps://integrate.api.nvidia.com/v1
ProtocolOpenAI-compatible chat completions
API keyNVIDIA_API_KEY
Model discoveryYes — the model list is asked of the endpoint rather than shipped.

Connecting it

The endpoint above is built in, so there is nothing to configure. Put the key in the environment and the provider appears in the model picker:

export NVIDIA_API_KEY="..."

Keys are issued by NVIDIA NIM — where to get one.

API key setup and troubleshooting

Official setup guide (checked 2026-10-24)

Create a key using the official guide below, then run this command and paste the key into the prompt. OHarness saves it in its credential store. A key set in the process environment takes precedence over a stored key.

oharness auth login nvidia

NVIDIA_API_KEY is an environment variable, not a top-level config.json field. For an endpoint override, merge the providers block below into ~/.oharness/config.json; preserve your other settings. The base URL must not include /chat/completions.

{
  "providers": {
    "nvidia": {
      "baseUrl": "https://integrate.api.nvidia.com/v1"
    }
  }
}

If the request fails

  • 401 or 403: read the response body, check the active key, account access and credit. An old environment variable can override a newly saved key. A terminal export only applies to that shell and processes started from it.
  • 404: check the endpoint and the exact model ID in the vendor console. 429: check the provider’s rate limits and quota before retrying. Model names in a bundled snapshot may no longer be available.

Install OHarness · Provider configuration · Configuration files

Models on NVIDIA NIM

The newest models NVIDIA NIM serves that can call tools, from the models.dev snapshot OHarness ships. The harness also asks the endpoint for its live list, so a model released after this snapshot still appears in the picker.

Model idContext / output$/M in / cached / outReasoning
z-ai/glm-5.3-flash1M / 131K0 / — / 0Yes
z-ai/glm-5.31M / 131K0 / — / 0Yes
deepseek-ai/deepseek-v4-pro-08131M / 384K0 / — / 0Yes
nvidia/nemotron-3.5-lightning-30b-a3b262K / 262K0 / — / 0Yes
meta/muse-glimmer-30b131K / 131K0 / — / 0Yes
deepseek-ai/deepseek-v4-flash-07311M / 384K0 / — / 0Yes
moonshotai/kimi-k31.05M / 131K0 / — / 0Yes
thinkingmachines/inkling1.05M / 16K0 / — / 0Yes
poolside/laguna-xs-2.1262K / 16K0 / — / 0Yes
z-ai/glm-5.21M / 131K0 / — / 0Yes

List prices in USD per million tokens, read from models.dev on 2026-09-19. NVIDIA NIM changes its prices and models; its own pricing page is the authority. Start a session on the newest:

oharness -m nvidia/z-ai/glm-5.3-flash

What you get either way

The same agent loop, the same tools, the same MCP servers and the same permission rules, whichever provider is underneath — and switching between them mid-session converts nothing, because history is never stored in a vendor’s shape. What an agent harness is covers why that is the hard part, and the providers reference covers the catalogue itself.

All 34 providers · Install OHarness