Problem

On Windows, memory-core plugin fails to load sqlite-vec extension, causing vector embeddings to not be generated. All memory writes show:

[memory] chunks written for ... without vector embeddings — chunks_vec not updated (sqlite-vec unavailable). Vector recall degraded for this file.

Environment

• OpenClaw: 2026.4.15
• OS: Windows (Windows_NT 10.0.26200)
• Node.js: v22.22.2
• sqlite-vec: installed at
  ode_modules/sqlite-vec-windows-x64/vec0.dll
• Ollama: installed and running with mxbai-embed-large model available

Expected behavior

sqlite-vec extension should be loaded successfully, enabling vector embedding generation for semantic memory search.

Actual behavior

• sqlite-vec-windows-x64 is installed (vec0.dll exists)
• Ollama with mxbai-embed-large is installed and responding
• But memory-core cannot load the sqlite-vec extension
• All memory writes fall back to FTS (keyword search only)

Diagnosis

The plugin SDK exports openMemoryDatabaseAtPath(dbPath, allowExtension) which accepts an �llowExtension parameter, but this is not exposed in the user configuration. The memory-core plugin appears to call this without the extension loading flag, or the Windows path resolution for vec0.dll fails.

What I tried

1. Ollama installed + mxbai-embed-large model pulled — working (verified via curl to localhost:11434)
2. sqlite-vec-windows-x64 package installed via npm — vec0.dll present
3. Config schema lookup shows no user-facing memorysearch or �llowExtension configuration path

px openclaw memory index still fails with sqlite-vec unavailable

Feature Request

Either:

1. Expose �llowExtension: true in memory-core config so users can enable it
2. Or ensure sqlite-vec loads automatically when the extension file is present
3. Or document the proper way to enable vector embeddings on Windows

Impact

• Vector recall degraded (semantic search broken)
• Dream/REM features may be impacted
• Users on Windows cannot use full memory capabilities even with Ollama installed