NovitaAI Provider Setup for OpenClaw

NovitaAI is a hosted AI infrastructure provider whose API is compatible with OpenAI. OpenClaw makes NovitaAI available via the official external @openclaw/novita-provider plugin. Model references take the novita/deepseek/deepseek-v4-pro form.

Setup

Install the plugin:

openclaw plugins install @openclaw/novita-provider

If a Gateway is running, installation takes effect on it right away; otherwise it applies at the next startup. See Apply changes and inspect.

Generate an API key at novita.ai/settings/key-management, then run:

openclaw onboard --auth-choice novita-api-key

Or set:

export NOVITA_API_KEY="<your-novita-api-key>" # pragma: allowlist secret

Defaults

Chat completions directed at the Novita endpoint rely on system messages and max_tokens, reasoning models included. This holds for catalog, configured, and discovered models alike, with no per-model compatibility settings needed.

Model catalog

• novita/moonshotai/kimi-k3
• novita/moonshotai/kimi-k2.7-code
• novita/minimax/minimax-m3
• novita/zai-org/glm-5.2
• novita/deepseek/deepseek-v4-pro
• novita/qwen/qwen3.7-max

novita/minimax/minimax-m2.7 can still be selected as a deprecated compatibility entry, though model pickers hide it.

Treat this as a starting point rather than a live catalog. Your account, your region, or Novita's current offering may add, remove, or restrict routes. Verify before you set a long-lived default:

openclaw models list --provider novita

Video generation

The video generation tool is supported by the same plugin and NOVITA_API_KEY, working through Novita's native asynchronous API at https://api.novita.ai.

When you supply one image to a -t2v model, the same family's -i2v route is selected. Remote image URLs and local image files are both accepted by these families, sent as data URIs. For Hailuo, 1080P works only with 6-second videos; pick 768P when the duration is 10 seconds. Video reference inputs are not supported.

Wan produces silent output by default; to generate audio, set audio: true. One remote HTTP(S) audio reference is also accepted. Text-to-video handles 16:9, 9:16, 1:1, 4:3, and 3:4; image-to-video keeps the source aspect ratio. Wan's options are negative_prompt, prompt_extend, shot_type (single or multi), and seed. Hailuo's options are enable_prompt_expansion and fast_pretreatment (the latter is not available on the Fast model).

{
  agents: {
    defaults: {
      mediaModels: {
        video: {
          primary: "novita/wan2.6-t2v",
        },
      },
    },
  },
}

When to choose Novita

• Access to hosted open-weight models through an OpenAI-compatible API.
• DeepSeek, Kimi, MiniMax, GLM, or Qwen-family routes under one provider account.
• An extra hosted fallback path alongside DeepInfra, GMI, OpenRouter, or direct vendor APIs.
• Model hosting handled by the provider, so you avoid running LM Studio, Ollama, SGLang, or vLLM infrastructure.

Pick a direct vendor provider if vendor-native request parameters or support contracts matter to you. Pick a local provider when the model has to run inside your own hardware or network boundary.

Troubleshooting

• 401/403: confirm the key in Novita's key management page, and re-run openclaw onboard --auth-choice novita-api-key when the stored profile is stale.
• Unknown model errors: use the exact novita/<route-id> that openclaw models list --provider novita returns.
• Slow or failed routes: switch to a different Novita model route, or configure Novita as a fallback provider for workloads that can tolerate provider-specific variance.

Related

• Model providers
• Provider directory