Memory

Buddy AI agents support two complementary kinds of memory:

Chat history — recent turns of the current conversation, replayed into the prompt so the agent has short-term context.
User memories — durable facts about a user, stored in a database and recalled across sessions.

Chat History (Short-Term)

The simplest form of memory needs no database. Set add_history_to_messages and control how many prior runs are included with num_history_runs.

from buddy import Agent
from buddy.models.openai import OpenAIChat

agent = Agent(
    name="chat_agent",
    model=OpenAIChat(id="gpt-4o-mini"),
    add_history_to_messages=True,
    num_history_runs=3,   # replay the last 3 exchanges
)

agent.print_response("My name is Alice.")
agent.print_response("What is my name?")   # remembers within the session

Reading chat history as a tool

Set read_chat_history=True to also give the agent a tool it can call to look back through the full conversation history on demand.

User Memories (Long-Term)

Long-term memories persist to a MemoryDb and are keyed by user_id, so they survive across sessions. Attach a Memory object and enable enable_user_memories on the agent.

from buddy import Agent
from buddy.models.openai import OpenAIChat
from buddy.memory.v2.memory import Memory
from buddy.memory.v2.db.sqlite import SqliteMemoryDb

agent = Agent(
    name="memory_agent",
    model=OpenAIChat(id="gpt-4o-mini"),
    memory=Memory(db=SqliteMemoryDb(table_name="memories", db_file="memory.db")),
    enable_user_memories=True,
)

agent.print_response("My name is Alice and I love hiking.", user_id="alice", session_id="s1")
agent.print_response("What do you know about me?",          user_id="alice", session_id="s2")

When enable_user_memories=True, the agent extracts salient facts after each run and stores them; on later runs they are retrieved and added to the context. Use enable_agentic_memory=True to instead let the agent decide when to create, update, or delete memories itself.

MemoryDb Backends

Memory stores user memories through a MemoryDb from buddy.memory.v2.db:

Backend	Class	Import
SQLite	`SqliteMemoryDb`	`buddy.memory.v2.db.sqlite`
PostgreSQL	`PostgresMemoryDb`	`buddy.memory.v2.db.postgres`
MongoDB	`MongoMemoryDb`	`buddy.memory.v2.db.mongodb`
Redis	`RedisMemoryDb`	`buddy.memory.v2.db.redis`
Firestore	`FirestoreMemoryDb`	`buddy.memory.v2.db.firestore`

Classic `AgentMemory`

The original AgentMemory (buddy.memory.agent) is still available and is also re-exported from the top level (from buddy import AgentMemory). It exposes flags such as create_user_memories, update_user_memories_after_run, and create_session_summary, and accepts a MemoryDb via its db field.

from buddy import Agent, AgentMemory
from buddy.models.openai import OpenAIChat
from buddy.memory.v2.db.sqlite import SqliteMemoryDb

agent = Agent(
    model=OpenAIChat(id="gpt-4o-mini"),
    memory=AgentMemory(
        db=SqliteMemoryDb(table_name="memories", db_file="memory.db"),
        create_user_memories=True,
        create_session_summary=True,
    ),
)

Session Storage

Chat history and run state can be persisted between process restarts by giving the agent storage (a buddy.storage backend such as SqliteStorage or PostgresStorage). See Configuration for storage setup.