Supported capabilities
Mem0
Mem0's memory layer: an LLM extracts facts from conversations, recalled by hybrid search, optionally reranked.
| Key | mem0 |
|---|
| From | PyPI |
|---|
| Installed | mem0ai==2.2.1, fastembed==0.9.1, spacy==3.8.18, en_core_web_sm-3.8.0-py3-none-any.whl |
|---|
| Licence | Apache-2.0 |
|---|
Versions
| Version | Upstream | Ports |
|---|
| Memory |
|---|
mem0 | 2.2.1 | ✓ |
Options
| Option |
Values, default in bold |
What it sets |
Versions |
embedder |
text-embedding-3-small, text-embedding-3-large |
The embedding model Mem0 indexes and searches with; Mem0's default is text-embedding-3-small. |
every version |
infer |
on, off |
Whether Mem0's LLM extracts facts from each session (on, Mem0's default) or the messages are stored as they are (off). |
every version |
reasoning_effort |
default, minimal, low, medium, high |
How much reasoning the model uses before answering, with default leaving the choice to the provider. |
every version |
rerank |
off, llm |
Whether search results are reranked by Mem0's LLM reranker, with the run's model (llm); Mem0 does not rerank by default. |
every version |
threshold |
0, 0.1, 0.3, 0.5 |
The lowest score a memory needs to be returned; Mem0's default is 0.1. |
every version |
top_k |
5, 10, 20, 50 |
How many memories a search returns at most; Mem0's default is 20. |
every version |
Benchmarks it fits