Chat History in Teams

Before running the examples, create and activate a virtual environment:

uv pip install agno openai sqlalchemy
export OPENAI_API_KEY="your-api-key"

Teams with storage enabled automatically have access to the run history of the session (also called the "conversation history" or "chat history").

We can give the Team access to the chat history in the following ways:

Team-Level History:

• You can set add_history_to_context=True and num_history_runs=5 to add the inputs and responses from the last 5 runs automatically to every request sent to the team leader.
• You can be more granular about how many messages to add to include in the list sent to the model, by setting num_history_messages.
• You can set read_chat_history=True to provide a get_chat_history() tool to your team allowing it to read any message in the entire chat history.
• You can enable search_past_sessions to allow searching through previous sessions.
• You can set add_team_history_to_members=True and num_team_history_runs=5 to add the inputs and responses from the last 5 runs (that is the team-level inputs and responses) automatically to every message sent to the team members.

Member-Level History:

• You can also enable add_history_to_context for individual team members. This will only add the inputs and outputs for that member to all requests sent to that member, giving it access to its own history.

The fragments below reuse this setup:

from agno.agent import Agent
from agno.db.sqlite import SqliteDb
from agno.team import Team

german_agent = Agent(name="German", instructions="Answer German questions.")
spanish_agent = Agent(name="Spanish", instructions="Answer Spanish questions.")
profile_agent = Agent(name="Profile", instructions="Explain profile settings.")
billing_agent = Agent(name="Billing", instructions="Explain billing concepts.")

History Reference

team = Team(
    members=[german_agent, spanish_agent],
    db=SqliteDb(db_file="tmp/team.db"),
    add_history_to_context=True,
    num_history_runs=5,
)

team = Team(
    members=[german_agent, spanish_agent],
    db=SqliteDb(db_file="tmp/team.db"),
    add_team_history_to_members=True,
    num_team_history_runs=3,
)

team = Team(
    members=[profile_agent, billing_agent],
    db=SqliteDb(db_file="tmp/team.db"),
    share_member_interactions=True,
)

team = Team(
    members=[german_agent, spanish_agent],
    db=SqliteDb(db_file="tmp/team.db"),
    read_chat_history=True,  # Team decides when to look up
)

team = Team(
    members=[german_agent, spanish_agent],
    db=SqliteDb(db_file="tmp/team.db"),
    search_past_sessions=True,
    num_past_sessions_to_search=10,
)

Add history to the team context

To add the history of the conversation to the context, you can set add_history_to_context=True. This will add the inputs and responses from the last 3 runs (that is the default) to the context of the team leader. You can change the number of runs by setting num_history_runs=n where n is the number of runs to include.

See the Direct Response with Team History example for a complete implementation.

Send team history to members

To send the team history to the members, you can set add_team_history_to_members=True. This will send the inputs and responses from the last 3 team-level runs (that is the default) to the members when tasks are delegated to them. You can change the number of runs by setting num_team_history_runs=n where n is the number of runs to include.

When enabled, team history is appended to the task sent to a team member in this format:

<team_history_context>
[run-1]
input: Hallo, wie heißt du? Mein Name ist John.
response: Ich heiße ChatGPT.

</team_history_context>

This allows members to access information from previous interactions with other team members.

See the Team History for Members example for a complete implementation.

Share member interactions with other members

All interactions with team members are automatically recorded, including the member name, the task given to the member, and the response from the member.

If you want members to have access to all interactions that has happened during the current run, you can set share_member_interactions=True.

When enabled, interaction details are appended to the task sent to a team member in this format:

<member_interaction_context>
See below interactions with other team members.
Member: Web Researcher
Task: Find information about the web
Response: I found information about the web

Member: HackerNews Researcher
Task: Find information about the web
Response: I found information about the web

</member_interaction_context>

See the Share Member Interactions example for a complete implementation.

Read the chat history

To read the chat history, you can set read_chat_history=True. This will provide a get_chat_history() tool to your team allowing it to read any message in the entire chat history.

Search the session history

In some scenarios, you might want to fetch messages from across multiple sessions to provide context or continuity in conversations.

Set search_past_sessions=True to give the team two tools: search_past_sessions() returns previews of recent sessions, and read_past_session(session_id) returns the full conversation for a specific session.

• num_past_sessions_to_search: Maximum number of past sessions to search. Defaults to 20.
• num_past_session_runs_in_search: Number of runs per session shown in the preview. Defaults to 3.

Developer Resources