Session Notebook

Source: src/langgraph_kit/core/memory/session.py

A thread-local structured notebook that maintains continuity within a conversation, especially useful across compaction events.

Class: SessionNotebook

Constructor

SessionNotebook(store: BaseStore, thread_id: str)

Stored at namespace ("session", thread_id), key "notebook".

Methods

Method	Description
initialize()	Create notebook from template if it doesn't exist
load()	Load current notebook content
save(content)	Overwrite entire notebook
update_section(name, content)	Replace content of a specific section
get_section(name)	Extract content of a specific section
should_update(messages)	Decide if notebook should update based on activity
estimate_tokens(content)	Rough token estimate (4 chars per token)
condense_section(name, max_tokens)	Truncate section if it exceeds budget
enforce_budget(max_total)	Condense sections if total exceeds budget

Notebook Sections

Section	Purpose
Current State	Where the conversation stands right now
Task Specification	What the user asked for
Files and Functions	Key files and code locations discussed
Workflow	Steps taken or planned
Errors and Corrections	Problems encountered and how they were resolved
Key Results	Important outputs or findings
Worklog	Chronological activity log

Update Thresholds

The notebook updates when activity exceeds thresholds:

Constant	Default	Trigger
DEFAULT_MESSAGE_THRESHOLD	6	Messages since last update
DEFAULT_TOOL_CALL_THRESHOLD	4	Tool calls since last update

Token Budgets

Constant	Default	Purpose
DEFAULT_MAX_SECTION_TOKENS	500	Max tokens per section
DEFAULT_MAX_TOTAL_TOKENS	3,000	Max tokens for entire notebook

When budgets are exceeded, enforce_budget() condenses sections by truncation.

Usage

The session notebook is primarily used by deep agents for maintaining context across long conversations:

notebook = SessionNotebook(store, thread_id)
await notebook.initialize()

# After significant work
await notebook.update_section("Current State", "Implementing auth middleware")
await notebook.update_section("Files and Functions", "- auth/middleware.py:42\n- tests/test_auth.py")

# Before compaction
content = await notebook.load()
# Content is included in compaction prompt for better summaries