Summary

The memory-lancedb plugin's openclaw.plugin.json schema restricts embedding.model to only two values:

"model": {
  "type": "string",
  "enum": ["text-embedding-3-small", "text-embedding-3-large"]
}

However, the TypeScript source (extensions/memory-lancedb/config.ts) already supports gemini-embedding-001 in the EMBEDDING_DIMENSIONS map:

const EMBEDDING_DIMENSIONS: Record<string, number> = {
  "text-embedding-3-small": 1536,
  "text-embedding-3-large": 3072,
  "gemini-embedding-001": 3072,  // <-- supported in code, rejected by schema
};

The config.ts also accepts baseUrl and apiKey in the embedding config, which allows using Google's OpenAI-compatible endpoint (https://generativelanguage.googleapis.com/v1beta/openai). But the JSON schema blocks baseUrl as an additional property.

Expected Behavior

The configSchema in openclaw.plugin.json should match the runtime capabilities:

"model": {
  "type": "string",
  "enum": ["text-embedding-3-small", "text-embedding-3-large", "gemini-embedding-001"]
}

And baseUrl should be added to the embedding schema properties:

"embedding": {
  "properties": {
    "apiKey": { "type": "string" },
    "model": { "type": "string", "enum": [...] },
    "baseUrl": { "type": "string" }
  }
}

Workaround

Currently must manually patch openclaw.plugin.json after every npm install / openclaw update, which is fragile.

Working Config (after patch)

{
  "plugins": {
    "entries": {
      "memory-lancedb": {
        "enabled": true,
        "config": {
          "embedding": {
            "apiKey": "<GEMINI_API_KEY>",
            "model": "gemini-embedding-001",
            "baseUrl": "https://generativelanguage.googleapis.com/v1beta/openai"
          },
          "autoCapture": true,
          "autoRecall": true
        }
      }
    },
    "slots": {
      "memory": "memory-lancedb"
    }
  }
}

Environment

• OpenClaw version: 2026.2.14
• OS: Ubuntu 24.04 (Hetzner VPS)
• Embedding provider: Google Gemini via OpenAI-compatible API