GMI Cloud Provider Setup for OpenClaw

GMI Cloud provides hosted inference for frontier and open-weight models through an API that is compatible with OpenAI. Within OpenClaw, it ships as an official external provider plugin: install it a single time, save credentials via the usual model auth flow, and reference models with refs such as gmi/openai/gpt-5.6-sol.

Pick GMI when a single API key should cover multiple hosted model families, including the Anthropic, DeepSeek, Google, Moonshot, OpenAI, and Z.AI routes that GMI's catalog exposes. It suits use as a secondary provider for fallback, for comparing hosted routes from different vendors, or when GMI carries a model before your primary provider does. The provider id, auth profile, aliases, model catalog seed, and base URL are managed by OpenClaw; live model availability, billing, rate limits, and any routing policy on the provider side are managed by GMI.

Setup

Install the plugin first, then generate an API key in GMI Cloud (https://www.gmicloud.ai/):

openclaw plugins install @openclaw/gmi-provider

When a Gateway is running, installation is applied to it automatically; if not, the change takes effect at the next startup. Refer to Apply changes and inspect.

After that, run:

openclaw onboard --auth-choice gmi-api-key

For installs driven by scripts or CI, supply every value on the command line:

openclaw onboard --non-interactive --accept-risk --skip-health \
  --auth-choice gmi-api-key \
  --gmi-api-key "$GMI_API_KEY"

Alternatively, put the key in the environment:

export GMI_API_KEY="<your-gmi-api-key>" # pragma: allowlist secret

When to choose GMI

• A hosted OpenAI-compatible endpoint is what you need, not a model server running locally.
• Several commercial and open-weight model families should be reachable through one provider account.
• You need a fallback provider whose upstream routing differs from DeepInfra, OpenRouter, Together, or the direct vendor APIs.
• GMI-specific model ids, pricing, or account controls are required.

Pick the direct vendor provider instead if you depend on vendor-native features that GMI's OpenAI-compatible route does not expose. Where data locality or local GPU control outweighs the convenience of hosting, choose a local provider such as LM Studio, Ollama, SGLang, or vLLM.

Models

For GPT-5.6, OpenAI tool turns automatically use reasoning_effort: "none"; for GPT-5.5 and GPT-5.4 mini that field is omitted.

The plugin catalog seeds the GMI Cloud route ids that are commonly available:

This catalog is a starting point, not a guarantee that every account can reach every model at any moment. To see what the configured provider reports in your environment, list the models:

openclaw models list --provider gmi

Troubleshooting

• 401 or 403: verify that GMI_API_KEY is set for the process running OpenClaw, or run onboarding again so the key is stored in the provider auth profile.
• Unknown model errors: check that the model exists in your GMI account, and use the full gmi/<route-id> ref that openclaw models list --provider gmi shows.
• Intermittent provider errors: switch to another GMI route, or set GMI up as a fallback instead of making it the sole primary model provider.

Related

• Model providers
• All providers