OpenClaw with free models from nvidia

If you are using OpenClaw quite intensively, you will find that the API cost might be a huge part of the bot. If you are trying to use top frontier models like Opus 4.6, GPT 5.3 Codex, you can easily burn hundreds/thousands of dollars per month. I tried to reduce operating cost by using some top-tier Chinese models, such as Kimi 2.5, Minimax 2.5, GLM 5 but it is hard to use the lowest subscription coding tier. For example, my Kimi Moderato account can only “live” for 2-3 days, and it burned all the weekly usage limit.

Another option is pushing some money to OpenRouter and using cheaper models, and limiting OpenClaw to run less frequently. But we have another option to try various models before paying: using NVIDIA Build API key. So this article will cover some notes to utilize OpenClaw with free models from Nvidia Build program.

1. First, register to https://build.nvidia.com/ to create an account
2. Then, generate API key at https://build.nvidia.com/settings/api-keys
3. Then, go to the Models menu and select the model that you are going to use. For example, I am using GLM5 from Z.ai, which was released some days ago.
4. Clicking on the “View Code” and we can see the base_url and model ID that we are going to use when configuring OpenClaw model.
5. Next, open your OpenClaw model configuration menu (openclaw configure > Models), choose “Custom Provider”:
   • In “API Base URL”, input base_url you get from NVIDIA model card: “https://integrate.api.nvidia.com/v1“
   • In “API Key (leave blank if not required)”, input YOUR_API_KEY
   • In “Endpoint compatibility”, choose “OpenAI-compatible (Uses /chat/completions)”
   • In “Model ID”, input the model ID you get from NVIDIA model card, e.g. “z-ai/glm5“
   • Wait for openclaw to verify and you are done
6. Next, we will need to change the context window for the model to work:
   • Open “~/.openclaw/openclaw.json“, search for the configuration of nvidia model (e.g. nvidia-glm5)
   • Change the proper value for “contextWindow“. This value depends on each model. For example with kimi-k2.5, it is “200000”
   • Change proper value for “maxTokens”, e.g. 8192
7. Done, restart gateway “openclaw gateway restart” and enjoy

I also attached a part of configured model in openclaw.json as follows.

"models": {
    "mode": "merge",
    "providers": {
      
      "nvidia-kimi": {
        "baseUrl": "https://integrate.api.nvidia.com/v1/",
        "apiKey": "nvapi-YOUR_API_KEY_HERE",
        "api": "openai-completions",
        "models": [
          {
            "id": "moonshotai/kimi-k2.5",
            "name": "NVIDIA Kimi K2.5",
            "reasoning": false,
            "input": [
              "text"
            ],
            "cost": {
              "input": 0,
              "output": 0,
              "cacheRead": 0,
              "cacheWrite": 0
            },
            "contextWindow": 256000,
            "maxTokens": 8192
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "nvidia-kimi/moonshotai/kimi-k2.5",
        "fallbacks": [
          "kimi-coding/k2p5"
        ]
      },
      "models": {
        "kimi-coding/k2p5": {
          "alias": "Kimi K2.5"
        },
        "nvidia-kimi/moonshotai/kimi-k2.5": {}
      },
    }
  },