openclaw-minimax-embeddings

OpenClaw memory embedding provider for MiniMax (model embo-01), wired to the existing minimax-portal OAuth profile. No new API keys required.

Why

OpenClaw's built-in memory embedding adapters all speak the OpenAI request format (input + input_type). MiniMax's embeddings API at https://api.minimax.io/v1/embeddings requires a different body shape (texts + type) and the type field is mandatory, so it can't be reused as-is. This plugin implements a small adapter that speaks MiniMax's protocol correctly.

Open feature request for upstream support: openclaw/openclaw#62065.

Install

# from the repo root
openclaw plugins install ./path/to/openclaw-minimax-embeddings
openclaw gateway restart

Or via npm once published:

openclaw plugins install @gotclaw/openclaw-minimax-embeddings
openclaw gateway restart

Configure

Add this to your openclaw.json:

{
  agents: {
    defaults: {
      memorySearch: {
        provider: "minimax",
        model: "embo-01",
      },
    },
  },
}

Optional overrides:

{
  agents: {
    defaults: {
      memorySearch: {
        provider: "minimax",
        model: "embo-01",
        // Override the endpoint (default: https://api.minimax.io/v1)
        remote: { baseUrl: "https://api.minimax.io/v1" },
        // Override type values
        queryInputType: "query",
        documentInputType: "db",
      },
    },
  },
}

Auth

The plugin resolves the bearer token through OpenClaw's auth system using the minimax-portal OAuth profile. No separate API key is needed — the OAuth access token is used directly, and resolveApiKeyForProvider handles automatic refresh via the stored refresh token before expires.

If the OAuth session is missing or revoked, the plugin's formatSetupError will return a clear message — re-run openclaw configure to repair the profile.

Verify

openclaw memory status --deep
openclaw memory index --force

Development

npm install
npm run build
npm run plugin:validate

License

MIT-0 (public domain dedication, no attribution required).