#Different memory search provider.

1 messages · Page 1 of 1 (latest)

ashen herald
#

I want to use an Ollama local embed model instad of the built in openai one because i do not pay for open ai.
when using memory search it fails and shows lots of warnings/timeouts in the gateway
i pulled and enabled mxbai 335m and enabled it as an available model
what now

small oysterBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> âś… Mark Solution

river prairieBOT
#

Memory search doesn’t use your “chat models” list — it has its own embedding config under agents.defaults.memorySearch. So “pulling + enabling mxbai 335m” was the right instinct, but it won’t be picked up automatically unless you point memorySearch at it.

Docs ref: https://docs.openclaw.ai/concepts/memory (Vector memory search section)

Option A (what you asked for): use Ollama’s embedding model via its OpenAI-compatible endpoint

Ollama exposes /v1/embeddings, so configure memorySearch as “openai”, but point it at your local Ollama base URL:

Config (JSON5):

agents: {
  defaults: {
    memorySearch: {
      enabled: true,
      provider: "openai",
      model: "mxbai-embed-large", // check `ollama list` for exact name
      fallback: "none",
# 
      remote: {
        baseUrl: "http://127.0.0.1:11434/v1/",
        apiKey: "ollama-local" // dummy is fine; Ollama doesn’t validate it
      }
    }
  }
}

Then restart the gateway. The index will rebuild automatically if provider/model/endpoint changed.

Common cause of your “timeouts/warnings”

The embedding calls happen from the Gateway machine. So if:

  • OpenClaw is in Docker and Ollama is on the host, 127.0.0.1 inside the container is not the host.
  • The gateway is on a VPS but Ollama is on your laptop, it also can’t reach it.