Store Namespaces

Complete inventory of LangGraph Store namespaces used by langgraph-kit.

All persistent data (beyond conversation messages in the checkpointer) is stored in the LangGraph Store under hierarchical tuple-based namespaces.

Namespace Map

Memory

Namespace	Key	Value	Module
("memory", scope, type)	record ID	MemoryRecord.to_store_value()	core/memory/persistent.py
("memory", "agent", agent_name, type)	record ID	MemoryRecord.to_store_value()	core/memory/agent_memory.py

Examples:

("memory", "user", "feedback")  → user feedback records
("memory", "project", "reference")  → project reference records
("memory", "agent", "researcher", "user")  → researcher's user knowledge

Session

Namespace	Key	Value	Module
("session", thread_id)	"notebook"	Notebook content string	core/memory/session.py

Orchestration

Namespace	Key	Value	Module
("async_tasks", parent_thread_id, task_id)	task_id	AsyncTask.model_dump()	core/orchestration/async_tasks.py
("queue", thread_id)	item ID	QueuedItem.model_dump()	core/orchestration/queue.py
("thread_busy",)	thread_id	{"busy": True, "timestamp": ...}	core/orchestration/queue.py

Tool Results

Namespace	Key	Value	Module
("tool_results",)	result ID	{"content": ..., "timestamp": ...}	core/context_management/result_persistence.py

Agent Backends (deepagents)

Namespace	Key	Value	Module
(agent_name, "memories")	item ID	Varies	core/graph_builder/backend.py
(agent_name, "notes")	item ID	Varies	core/graph_builder/backend.py

Pruning

The pruning.py module cleans up stale data from these namespaces:

Namespace	Pruning Rule
("tool_results",)	Items older than max_age_seconds (default 7 days)
("thread_busy",)	Locks older than 600 seconds (auto-expire)

from langgraph_kit.pruning import prune_store

result = await prune_store(store, max_age_seconds=86400 * 7)
print(f"Deleted {result.tool_results_deleted} old tool results")
print(f"Cleared {result.stale_locks_cleared} stale locks")