Skip to main content
OpenClaw manages models through providers. When you connect it to the powertokens platform, choose the protocol path first:
  • OpenAI-compatible path: use openai-completions with https://api.powertokens.ai/v1.
  • Anthropic path: use anthropic-messages with https://api.powertokens.ai.
Do not conflate OpenClaw’s official OpenAI/Codex runtime, Claude CLI reuse, and a regular OpenAI-compatible proxy. This guide covers the powertokens platform as a custom provider. If you need to choose the endpoint first, see Text model protocols and endpoints.

​
Choose a path

GoalOpenClaw provider apiBase URLModel scope
OpenAI-compatible text modelsopenai-completionshttps://api.powertokens.ai/v1Text models that support /v1/chat/completions
Anthropic-path modelsanthropic-messageshttps://api.powertokens.aiRecommended Zhipu, MiniMax, Qwen, BytePlus DeepSeek, and re-validated Seed models that support Messages
OpenAI-compatible preferred modelsopenai-completionshttps://api.powertokens.ai/v1Text models whose upstream provider does not support /v1/messages, such as seed-1-6-flash-250715

​
OpenAI-compatible provider example

Add a provider like this to OpenClaw models.providers. The actual location can be global config or the target agent’s models.json. 
{
  models: {
    mode: "merge",
    providers: {
      powertokens: {
        baseUrl: "https://api.powertokens.ai/v1",
        apiKey: "YOUR_POWERTOKENS_API_KEY",
        api: "openai-completions",
        models: [
          {
            id: "glm-5.2",
            name: "glm-5.2",
            reasoning: false,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            contextWindow: 128000,
            maxTokens: 32000
          },
          {
            id: "seed-2-0-pro-260328",
            name: "seed-2-0-pro-260328",
            reasoning: false,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            contextWindow: 128000,
            maxTokens: 32000
          }
        ]
      }
    }
  },
  agents: {
    defaults: {
      models: {
        "powertokens/glm-5.2": { alias: "powertokens-glm" },
        "powertokens/seed-2-0-pro-260328": { alias: "powertokens-seed-pro" }
      }
    }
  }
}
Do not commit a real API key. Use OpenClaw-supported environment, SecretRef, or local secret configuration for production.

​
Anthropic provider example

For the Claude/Anthropic path, set api to anthropic-messages and use the base URL without /v1/messages. 
{
  models: {
    mode: "merge",
    providers: {
      "powertokens-anthropic": {
        baseUrl: "https://api.powertokens.ai",
        apiKey: "YOUR_POWERTOKENS_API_KEY",
        api: "anthropic-messages",
        models: [
          {
            id: "glm-5.2",
            name: "glm-5.2",
            reasoning: false,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            contextWindow: 128000,
            maxTokens: 32000
          },
          {
            id: "MiniMax-M3",
            name: "MiniMax-M3",
            reasoning: false,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            contextWindow: 128000,
            maxTokens: 32000
          },
          {
            id: "deepseek-v3-2-251201",
            name: "deepseek-v3-2-251201",
            reasoning: false,
            input: ["text"],
            cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
            contextWindow: 128000,
            maxTokens: 32000
          }
        ]
      }
    }
  },
  agents: {
    defaults: {
      models: {
        "powertokens-anthropic/glm-5.2": { alias: "powertokens-claude-glm" },
        "powertokens-anthropic/MiniMax-M3": { alias: "powertokens-claude-minimax" },
        "powertokens-anthropic/deepseek-v3-2-251201": { alias: "powertokens-claude-deepseek" }
      }
    }
  }
}
Recommended Anthropic tool-workflow models:
Upstream providerModels
Zhipuglm-4.5-air, glm-4.7, glm-4.7-flash, glm-5, glm-5-turbo, glm-5.2
MiniMaxMiniMax-M2.5, MiniMax-M2.5-highspeed, MiniMax-M2.7, MiniMax-M2.7-highspeed, MiniMax-M3
Alibaba Qwenqwen3-coder-plus, qwen3-max, qwen3.5-flash, qwen3.6-plus
BytePlus DeepSeekdeepseek-v3-2-251201
BytePlus Seedseed-1-6-250915, seed-1-8-251228, seed-2-0-lite-260228, seed-2-0-mini-260215, seed-2-0-pro-260328
Use these with openai-completions, not anthropic-messages, even if model discovery declares anthropic: 
seed-1-6-flash-250715

​
Apply and verify

After applying config for your OpenClaw installation, run: 
openclaw doctor
openclaw models list
openclaw models status --json
openclaw gateway restart
Then select the new model in TUI or a connected channel and send: 
Reply with ok, then list the active model name.
If the model is unavailable, check:
  • The model reference uses provider/model.
  • models.providers.*.api matches the platform path.
  • baseUrl has the correct /v1 shape for the selected protocol.
  • The API key is injected through the OpenClaw secret mechanism you use.
  • The model is in the correct OpenAI-compatible or Anthropic-path list.

​
Boundaries

  • openai-completions covers Chat Completions semantics. It is not the same as OpenClaw’s official OpenAI/Codex runtime.
  • anthropic-messages only applies to models that work through /v1/messages.
  • OpenClaw tools, media, caching, reasoning, and developer-role compatibility may depend on provider metadata. Validate with real tasks.
  • If the anthropic-messages path is unstable in real tasks, switch to openai-completions and keep the request id for platform debugging.
  • Keep input: ["text"] for text-only models. Do not declare image support unless the model and platform API support it.