RecencyMemory

Recency-based episodic memory.

RecencyMemory

Bases: BaseMemory

Recency-based episodic memory.

Recalls the last N episodes from the store and formats them for prompt injection. The simplest possible memory implementation: append on write, last-N on recall.

Attributes:

Name	Type	Description
store	BaseMemoryStore	The underlying memory store.
n	int	Number of most recent episodes to recall.

Source code in src/llm_agents_from_scratch/memory/recency.py

class RecencyMemory(BaseMemory):
    """Recency-based episodic memory.

    Recalls the last N episodes from the store and formats them for prompt
    injection. The simplest possible memory implementation: append on write,
    last-N on recall.

    Attributes:
        store (BaseMemoryStore): The underlying memory store.
        n (int): Number of most recent episodes to recall.
    """

    def __init__(self, store: BaseMemoryStore, n: int = 3) -> None:
        """Initialize a RecencyMemory.

        Args:
            store (BaseMemoryStore): The memory store to read from and write
                to.
            n (int): Number of most recent episodes to include in recall.
                Defaults to 3.
        """
        self.store = store
        self.n = n

    async def recall(self, task: Task) -> str:
        """Retrieve the N most recent episodes as a formatted string.

        Args:
            task (Task): The incoming task (unused — recency recall is
                unconditional).

        Returns:
            str: Newline-separated episode strings, or empty string if the
                store is empty.
        """
        episodes = await self.store.read_recent(self.n)
        if not episodes:
            return ""
        return "\n".join(str(ep) for ep in episodes)

    async def record(self, episode: Episode) -> None:
        """Persist a completed episode to the store.

        Args:
            episode (Episode): The completed episode to store.
        """
        await self.store.write(episode)

    async def summary(self) -> str:
        """Return a human-readable summary of the memory and its store.

        Describes the recall strategy (last N), then delegates to the
        store for substrate-level facts (count, oldest, newest).

        Returns:
            str: Multi-line summary of the memory and its store.
        """
        store_summary = await self.store.summary()
        return f"RecencyMemory (recall last {self.n}):\n{store_summary}"

__init__

__init__(store, n=3)

Initialize a RecencyMemory.

Parameters:

Name	Type	Description	Default
store	BaseMemoryStore	The memory store to read from and write to.	required
n	int	Number of most recent episodes to include in recall. Defaults to 3.	3

Source code in src/llm_agents_from_scratch/memory/recency.py

def __init__(self, store: BaseMemoryStore, n: int = 3) -> None:
    """Initialize a RecencyMemory.

    Args:
        store (BaseMemoryStore): The memory store to read from and write
            to.
        n (int): Number of most recent episodes to include in recall.
            Defaults to 3.
    """
    self.store = store
    self.n = n

recall async

recall(task)

Retrieve the N most recent episodes as a formatted string.

Parameters:

Name	Type	Description	Default
task	Task	The incoming task (unused — recency recall is unconditional).	required

Returns:

Name	Type	Description
str	str	Newline-separated episode strings, or empty string if the store is empty.

Source code in src/llm_agents_from_scratch/memory/recency.py

async def recall(self, task: Task) -> str:
    """Retrieve the N most recent episodes as a formatted string.

    Args:
        task (Task): The incoming task (unused — recency recall is
            unconditional).

    Returns:
        str: Newline-separated episode strings, or empty string if the
            store is empty.
    """
    episodes = await self.store.read_recent(self.n)
    if not episodes:
        return ""
    return "\n".join(str(ep) for ep in episodes)

record async

record(episode)

Persist a completed episode to the store.

Parameters:

Name	Type	Description	Default
episode	Episode	The completed episode to store.	required

Source code in src/llm_agents_from_scratch/memory/recency.py

async def record(self, episode: Episode) -> None:
    """Persist a completed episode to the store.

    Args:
        episode (Episode): The completed episode to store.
    """
    await self.store.write(episode)

summary async

summary()

Return a human-readable summary of the memory and its store.

Describes the recall strategy (last N), then delegates to the store for substrate-level facts (count, oldest, newest).

Returns:

Name	Type	Description
str	str	Multi-line summary of the memory and its store.

Source code in src/llm_agents_from_scratch/memory/recency.py

async def summary(self) -> str:
    """Return a human-readable summary of the memory and its store.

    Describes the recall strategy (last N), then delegates to the
    store for substrate-level facts (count, oldest, newest).

    Returns:
        str: Multi-line summary of the memory and its store.
    """
    store_summary = await self.store.summary()
    return f"RecencyMemory (recall last {self.n}):\n{store_summary}"