Sessions#

An agent is designed once, and when it is deployed there is not a copy of it for each user. There is a single agent instance. Therefore, we need a way to organize each user context. Of course, at the same time there can be multiple users, each of them in a different agent state, with different values for some variables, etc. To handle this, we introduce the Session object.

A session is an object assigned to each user where it is stored private data.

If you dig into the BAF’s guts, you will see that some functions receive a Session object as a parameter. That is because these functions can modify the user session. Most of the session processes are done internally, so you don’t need to worry about it.

You are able to to manipulate the user session in 2 places:

  • State body functions (see the State body documentation)

def example_body(session: Session):
    # Your code here

  • Transition conditions (see the Conditions documentation)

def example_condition(session: Session, params: dict):
    # Your code here

We will stick to the body use case, although the same can be done in transition conditions.

To understand how this internally works, we must know that the body function is defined once, and assigned to a state. Whenever a user moves to a state, this function is executed by the agent, but taking as argument the user session. This way the agent can read and write user-specific data stored there and it cannot do it with other users data.

Let’s see the different things we can do with a user session.

(here we specify the type of each object, although it is not necessary to do it)

def example_body(session: Session):
    # We can get the received event:
    event: Event = session.event  # Each type of event has different information we can access
    # We can get the session id:
    session_id: str =  session.id
    # We can send a message to the user:
    session.reply('Hello!')
    # We can get the chat history (the 'n' last messages):
    chat_history: list[Message] = session.get_chat_history(n=5)
    # We can set (store) a variable:
    session.set('age', 30)
    # We can get a variable (the return type can be any type):
    age: int = session.get('age')
    # We can delete a variable:
    session.delete('age')
    # We can send a message to another agent
    session.send_message_to_websocket('ws://localhost:5000', 'Hello!')
    # We can run RAG
    session.run_rag('your question')

Note

See Events to learn about Events

Note

To access the chat history, you need to set up the Monitoring Database

Note

See Communication between agents: Multi-agent systems to learn about communication between agents

Note

See RAG to learn about Retrieval Augmented Generation

API References#