Session Context

Goals, plans, and progress for active sessions.

The Session Context Store captures the current state of a conversation: what’s been discussed, what the goal is, and what progress has been made. Unlike other stores that accumulate data, session context is a snapshot that gets replaced on each update.

Aspect	Value
Scope	Per session
Persistence	Session lifetime (replaced on update)
Default mode	Always
Supported modes	Always

Basic Usage

from agno.agent import Agent
from agno.db.postgres import PostgresDb
from agno.learn import LearningMachine
from agno.models.openai import OpenAIResponses

agent = Agent(
    model=OpenAIResponses(id="gpt-5.2"),
    db=PostgresDb(db_url="postgresql+psycopg://ai:ai@localhost:5532/ai"),
    learning=LearningMachine(session_context=True),
)

# Session tracks what's being discussed
agent.print_response(
    "I'm designing a REST API for a todo app. Should I use PUT or PATCH for updates?",
    user_id="[e-mail]",
    session_id="api_design",
)

# Later in the session, context is maintained
agent.print_response(
    "What about the delete endpoint?",
    user_id="[e-mail]",
    session_id="api_design",
)

The agent knows the ongoing context about REST API design.

Summary Mode

Default behavior. Captures the essence of the conversation without detailed planning.

from agno.learn import LearningMachine, SessionContextConfig

agent = Agent(
    model=OpenAIResponses(id="gpt-5.2"),
    db=db,
    learning=LearningMachine(
        session_context=SessionContextConfig(),
    ),
)

What gets captured: what’s being worked on, key decisions made, current state, open questions.

Planning Mode

Enable planning to track goals, plan steps, and progress.

from agno.learn import LearningMachine, SessionContextConfig

agent = Agent(
    model=OpenAIResponses(id="gpt-5.2"),
    db=db,
    learning=LearningMachine(
        session_context=SessionContextConfig(enable_planning=True),
    ),
)

agent.print_response(
    "Help me deploy a Python app to production. Give me the steps.",
    user_id="[e-mail]",
    session_id="deploy_app",
)

# Later, progress is tracked
agent.print_response(
    "Done with step 1. What's next?",
    user_id="[e-mail]",
    session_id="deploy_app",
)

Data Model

Field	Description
session_id	Unique session identifier
user_id	User this session belongs to
summary	What’s been discussed
goal	What user is trying to accomplish (planning mode)
plan	Steps to achieve goal (planning mode)
progress	Completed steps (planning mode)
created_at	When created
updated_at	Last update

Accessing Session Context

lm = agent.get_learning_machine()

context = lm.session_context_store.get(session_id="api_design")
if context:
    print(f"Summary: {context.summary}")
    if context.goal:
        print(f"Goal: {context.goal}")

# Debug output
lm.session_context_store.print(session_id="api_design")

Context Injection

Session context is injected into the system prompt:

<session_context>
Summary: Helping user design a REST API for a todo app. Discussed resource naming conventions. Currently exploring HTTP methods for CRUD operations.

Goal: Design complete REST API for todo application

Plan:
  1. Define resource endpoints
  2. Choose HTTP methods for each operation
  3. Design request/response schemas
  4. Add authentication

Completed:
  ✓ Define resource endpoints
</session_context>

When to Use

Session context is essential when:

• Message history gets truncated: long conversations lose early context
• Sessions are resumed: user returns after a break
• Complex multi-step tasks: track progress through long workflows
• Handoffs: another agent or human needs to understand the state

Combining with Other Stores

Session context works well alongside user-level stores:

from agno.learn import LearningMachine, SessionContextConfig

agent = Agent(
    model=OpenAIResponses(id="gpt-5.2"),
    db=db,
    learning=LearningMachine(
        user_profile=True,                                  # Who the user is
        session_context=SessionContextConfig(enable_planning=True),  # Current state
    ),
)

Long-term user knowledge plus short-term session state.