Bases: Memory
Long-term memory kept by the mem0 cloud service, searched by meaning.
Requires MEM0_API_KEY in the environment, or pass api_key explicitly.
Memories are kept per ctx.user_id, or per ctx.session_id when there is no
user.
Example
from ant_ai.agent import Agent
from ant_ai.memory.backends.mem0 import Mem0Memory
agent = Agent(memory=Mem0Memory(), ...)
The agent can now call:
- Mem0Memory_search(query="user preferences")
- Mem0Memory_add(facts=["The user prefers answers in Italian"])
Fields:
Validators:
Source code in src/ant_ai/memory/backends/mem0.py
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93 | class Mem0Memory(Memory):
"""
Long-term memory kept by the mem0 cloud service, searched by meaning.
Requires MEM0_API_KEY in the environment, or pass ``api_key`` explicitly.
Memories are kept per `ctx.user_id`, or per `ctx.session_id` when there is no
user.
Example:
```python
from ant_ai.agent import Agent
from ant_ai.memory.backends.mem0 import Mem0Memory
agent = Agent(memory=Mem0Memory(), ...)
```
The agent can now call:
- Mem0Memory_search(query="user preferences")
- Mem0Memory_add(facts=["The user prefers answers in Italian"])
"""
api_key: str | None = Field(default=None)
top_k: int = Field(default=5, ge=1, description="Most memories one search returns.")
_client: Any = PrivateAttr()
@model_validator(mode="after")
def _init_client(self) -> Mem0Memory:
self._client: AsyncMemoryClient = (
AsyncMemoryClient(api_key=self.api_key)
if self.api_key
else AsyncMemoryClient()
)
return self
async def search(
self, query: str, ctx: InvocationContext | None = None
) -> list[str]:
"""Search long-term memory for facts relevant to `query`. Call this
whenever recalling something about the user or a past conversation
would help answer the current request."""
filters: dict[str, str] = _resolve_scope(ctx)
options = SearchMemoryOptions(filters=filters, top_k=self.top_k)
results: Any = await self._client.search(query, options=options)
return [r["memory"] for r in results.get("results", []) if r.get("memory")]
async def add(self, facts: list[str], ctx: InvocationContext | None = None) -> str:
"""Persist one or more durable facts (user preferences, personal
details, explicit "remember this" instructions) for future
conversations. Call this proactively as soon as you learn something
worth keeping — do not wait to be asked, and do not wait until the
end of the conversation."""
facts = [f for f in facts if f]
if not facts:
return "Nothing to save."
filters: dict[str, str] = _resolve_scope(ctx)
await self._client.add(
[{"role": "system", "content": f} for f in facts], **filters
)
return f"Saved {len(facts)} fact(s) to memory."
|
search
async
Search long-term memory for facts relevant to query. Call this
whenever recalling something about the user or a past conversation
would help answer the current request.
Source code in src/ant_ai/memory/backends/mem0.py
69
70
71
72
73
74
75
76
77
78 | async def search(
self, query: str, ctx: InvocationContext | None = None
) -> list[str]:
"""Search long-term memory for facts relevant to `query`. Call this
whenever recalling something about the user or a past conversation
would help answer the current request."""
filters: dict[str, str] = _resolve_scope(ctx)
options = SearchMemoryOptions(filters=filters, top_k=self.top_k)
results: Any = await self._client.search(query, options=options)
return [r["memory"] for r in results.get("results", []) if r.get("memory")]
|
add
async
Persist one or more durable facts (user preferences, personal
details, explicit "remember this" instructions) for future
conversations. Call this proactively as soon as you learn something
worth keeping — do not wait to be asked, and do not wait until the
end of the conversation.
Source code in src/ant_ai/memory/backends/mem0.py
80
81
82
83
84
85
86
87
88
89
90
91
92
93 | async def add(self, facts: list[str], ctx: InvocationContext | None = None) -> str:
"""Persist one or more durable facts (user preferences, personal
details, explicit "remember this" instructions) for future
conversations. Call this proactively as soon as you learn something
worth keeping — do not wait to be asked, and do not wait until the
end of the conversation."""
facts = [f for f in facts if f]
if not facts:
return "Nothing to save."
filters: dict[str, str] = _resolve_scope(ctx)
await self._client.add(
[{"role": "system", "content": f} for f in facts], **filters
)
return f"Saved {len(facts)} fact(s) to memory."
|