Problem

The memory-lancedb long-term memory plugin requires an OpenAI API key (embedding.apiKey) and only supports text-embedding-3-small or text-embedding-3-large. There is no way to use it with local embedding providers like Ollama.

This creates a hard dependency on a paid external service for a feature that could otherwise run entirely locally. Agents who use local embedding stacks (Ollama, llama.cpp, etc.) are excluded.

Impact

• Agents on self-hosted setups without OpenAI keys cannot use memory-lancedb at all
• Projects like ClawBrain already solve local vector memory using Ollama — the bundled plugin should be at least as flexible
• Forces cost on agents who want persistent memory but don't want to pay per-embedding

Proposed Solution

Support pluggable embedding providers in memory-lancedb:

{
  "plugins": {
    "slots": { "memory": "memory-lancedb" },
    "entries": {
      "memory-lancedb": {
        "config": {
          "embedding": {
            "provider": "ollama",
            "model": "all-minilm",
            "url": "http://localhost:11434"
          }
        }
      }
    }
  }
}

Or at minimum, support OpenAI-compatible endpoints so agents can point it at a local Ollama instance running an embedding model.

Current workaround

Use memory-core (bundled, no external key needed) or an external vector memory system like ClawBrain that uses Ollama natively.