[Question]: OpenClaw + OpenViking Memory Extraction Issue (0 Memories Returned)

[doremo2021]

Your Question

Hi team,

I’m encountering a memory extraction issue between OpenClaw and OpenViking, which are running on separate servers.

Setup

• OpenClaw: agent framework (running on a separate server)

• OpenViking: memory server

• Ollama (local on OpenViking server)

  • qwen2.5:3b → used for memory extraction (LLM)
  • qwen3-embedding:0.6b → used for embeddings

Verification
I tested the models directly using the Ollama API:

• qwen2.5:3b via /api/chat → works and returns responses normally
• qwen3-embedding:0.6b via /v1/embeddings → works and returns vectors
• OpenViking can reach the local Ollama endpoint (127.0.0.1:11434)

So from the API side, both chat and embedding appear to be functioning correctly.

Problem
When OpenClaw sends memory to OpenViking, the logs show:

memory_store completed but extract returned 0 memories

and in the OpenViking service logs:

Memory extraction failed: Connection error

This seems to indicate the memory extractor pipeline fails before returning extracted memories, even though the models themselves are reachable and working when tested directly.

Any suggestions or debugging tips would be appreciated.

Thanks!