← v2026.3.2

Ollama Memory Embeddings

v2026.3.2 · Release notes

Use local Ollama for memory search. Set `memorySearch.provider = "ollama"` for fully local, private memory — no API calls.

Ollama for memory embeddings

You can now use local Ollama for memory search — no API calls, no cloud, fully private.

Configuration

{
  "agents": {
    "defaults": {
      "memorySearch": {
        "provider": "ollama",
        "fallback": "ollama"
      }
    }
  }
}

Ollama uses your existing models.providers.ollama settings. Make sure you have an embedding model running (e.g. nomic-embed-text).

Why use Ollama for memory?

  • Privacy — Your memories never leave your machine
  • Cost — No embedding API fees
  • Offline — Works without internet
  • Speed — Local inference, no network latency

Fallback

If you set fallback: "ollama", OpenClaw will try the primary provider first and fall back to Ollama on failure.