John Lockwood
John Lockwood
Pydantic is already being used for config: `MemoryConfig` in the `Memory` and `AsyncMemory` constructor. https://github.com/mem0ai/mem0/blob/ec1d7a45d3d0a11894a16cd7d05d6ba0257141f6/mem0/memory/main.py#L41 https://github.com/mem0ai/mem0/blob/ec1d7a45d3d0a11894a16cd7d05d6ba0257141f6/mem0/configs/base.py#L29 the `Mem0` class constructor could be changed so that if `config` is an instance...
I attempted to construct a MemoryConfig with non-dict configs all the way down, but the sub-configs are not designed for that. ``` llm_config = LlmConfig( provider="openai", config=BaseLlmConfig( model="qwen/qwen3-30b-a3b:free", api_key=os.environ.get("OPENROUTER_API_KEY"), openrouter_base_url="https://openrouter.ai/api/v1",...
This change works and now persists memory between container down/up. with: ``` volumes: - mem0_storage:/mem0/storage ``` /qdrant/storage is not on the volume with: ``` volumes: - mem0_storage:/qdrant/storage ``` /qdrant/storage is...
The [docs](https://docs.nomic.ai/atlas/embeddings-and-retrieval/text-embedding) for the embedding model `nomic-embed-text:latest` shows it's dims are 768. So config the vector_store's embedding_model_dims to 768. ``` "vector_store": { "provider": "qdrant", "config": { "collection_name": "test", "host": "localhost",...
It would probably be easier to review the feature changes without format changes to otherwise unchanged files. I recommend installing `ruff` and setting ruff as the formatter in your editor....
[existing MCP issue](https://github.com/modelcontextprotocol/python-sdk/issues/423) Try restarting the client after restarting openMemory. I just had to quit Cursor and reopen and it worked.
While you wait on this PR, you may configure mem0/Memory with a custom_update_memory_prompt. ``` config: MemoryConfig = { "vector_store": { ... }, "custom_update_memory_prompt": "< your custom prompt >" } memory...
I don't have the issue with `...` in the response from the llm anymore, I have a feeling that I had an out-of-date openai client installed. At some point I...
You are right that `metadata` isn't handled in a way that is safe for concurrency/thread safe, not sure if `new_message_embeddings` is either, because it's mutated in the process_fact inner function....