Configure AIFast in OpenClaw

OpenClaw can be accessed via models.providers Add a custom OpenAI Compatible provider. Do not directly override the built-in openai Provider; use a separate provider ID for AIFast so that subsequent troubleshooting and switching will be clearer.

Link provider configuration and plan scope in OpenClaw

Keep a separate provider ID for Link, with its key, base URL and model mapping together. GPT text, Claude Messages and image generation require different capability checks. A model appearing in the directory does not justify declaring untested tool or image support.

Before running an agent on a Link plan, check the key group, daily allowance and wallet-fallback setting. Inspect the request log for a short task before enabling concurrency, heartbeats or batches, so background loops do not exhaust both the allowance and wallet.

Review Link funding and monthly-plan rules

Quick conclusion

Create a dedicated AIFast provider. The model ID must come from the current /v1/models response and select the adapter that matches the target model's actual protocol. A saved configuration only proves that the local file parses; verify the destination with separate minimal text, streaming, and tool-call requests.

Check before you start

1. Request using AIFast Key /v1/models, copying the real model ID.
2. Confirm that the target model uses Chat Completions or Responses API.
3. Create an independent API Key for OpenClaw and save it as an environment variable.

export AIFAST_API_KEY="your_API_Key"

Configuration steps

Incorporate an independent provider in the OpenClaw configuration. The following example is for support /v1/chat/completions model:

{
  models: {
    mode: "merge",
    providers: {
      aifast: {
        baseUrl: "https://www.aifast.link/v1",
        apiKey: "${AIFAST_API_KEY}",
        api: "openai-completions",
        models: [
          {
            id: "YOUR_MODEL_ID",
            name: "AIFast station model"
          }
        ]
      }
    }
  },
  agents: {
    defaults: {
      model: {
        primary: "aifast/YOUR_MODEL_ID"
      }
    }
  }
}

place two YOUR_MODEL_ID Replace with /v1/models The same model ID returned.

Choose the right API adapter

Only actually supported in the target model and compatibility layer /v1/responses Use only when openai-responses. Don't judge based on the model name alone.

Verify whether the model is registered

Confirm using OpenClaw's model list or status command after configuration:

• provider is shown as aifast.
• The model route appears as aifast/YOUR_MODEL_ID.
• The final request address does not appear /v1/v1.

Then run a short task that does not involve sensitive files and confirm that normal text, streaming output, and tool calls are available respectively.

FAQ

The model is displayed, but the request returns model not found

The successful registration of the OpenClaw local model directory does not mean that the upstream accepts the model name. Call AIFast again /v1/models, check the case, prefix and account open status.

Ordinary chat is successful, but tool invocation fails.

Basic text requests only prove that Chat Completions are available. Tool calls also rely on model capabilities, compatibility layer event formats, and OpenClaw adapters. Save the complete error text first, and do not blindly switch prompt words.

Still using old model after configuration

Confirm agents.defaults.model.primary pointed to aifast/YOUR_MODEL_ID, and check whether the current Agent has a separate model override configuration.

Next step

After integration, give OpenClaw a separate key and spending limit. Test read-only tools, file-writing tools, long-running tasks, and recovery from 429 rate limits. Review tool authorization separately when browser, shell, or file permissions are involved.