Add mem0 as a memory provider #3129
Conversation
|
@parshvadaftari Thanks for working on this. Did you see #2932 though? I was expecting a much more lightweight architecture built on top of toolsets, rather than requiring a completely new module and abstract classes. Can you see if you could reimagine this as |
|
Here's mine implementation for reference. https://gist.github.com/nurikk/71f80d8fa375c2a038c6e8e81ab8b6a9 |
|
Hey @DouweM Have implemented the changes and made sure the tests passed. |
| allowing agents to save and search through conversation memories. | ||
| """ | ||
|
|
||
| # pyright: reportUnknownMemberType=false, reportUnknownArgumentType=false, reportUnknownVariableType=false |
There was a problem hiding this comment.
Please remove this, we have these rules for a reason!
| from .function import FunctionToolset | ||
|
|
||
| if TYPE_CHECKING: | ||
| from mem0 import AsyncMemoryClient, MemoryClient |
There was a problem hiding this comment.
I don't think we need it if we already have it below
| AsyncMemoryClient = None # pragma: no cover | ||
| MemoryClient = None # pragma: no cover | ||
| else: | ||
| _import_error = None |
There was a problem hiding this comment.
We can show the user new "please install..." error right away
|
|
||
| This toolset adds two tools to your agent: | ||
| - `save_memory`: Save information to memory for later retrieval | ||
| - `search_memory`: Search through stored memories |
There was a problem hiding this comment.
Those docs have weird names like _search_memory_impl
| if client is not None: | ||
| self.client = client | ||
| # Check if client is async by looking for async methods | ||
| self._is_async = hasattr(client, 'search') and hasattr(getattr(client, 'search'), '__call__') |
There was a problem hiding this comment.
Is this the best way to do this? Could we not check isinstance?
| self._is_async = inspect.iscoroutinefunction(client.search) | ||
| else: | ||
| # Create async client by default for better performance | ||
| self.client = AsyncMemoryClient( # type: ignore[misc] |
There was a problem hiding this comment.
What error are we ignoring here?
|
|
||
| return '\n'.join(memory_lines) | ||
|
|
||
| except Exception as e: |
There was a problem hiding this comment.
Can we be more specific in the errors we mask?
| else: | ||
| response = self.client.search(query=query, user_id=user_id, limit=self._limit) | ||
|
|
||
| # Parse response - handle both dict format and raw list |
There was a problem hiding this comment.
Why? Examples like https://docs.mem0.ai/integrations/google-ai-adk treat it as a list
| memory_lines = ['Found relevant memories:'] | ||
| for mem in results: | ||
| if isinstance(mem, dict): | ||
| memory_text = mem.get('memory', '') |
There was a problem hiding this comment.
If the memory object is invalid, we should just skip it right
| return 'No relevant memories found.' | ||
|
|
||
| # Format memories for the agent | ||
| memory_lines = ['Found relevant memories:'] |
There was a problem hiding this comment.
Do we need this prefix? The examples in their docs don't use it
This integration introduces memory layer for the AI agents, leveraging mem0 as primary memory provider.