knowledge_graph

Knowledge Graph Operations

The user's knowledge is organized in a personal knowledge graph. Each concept is a node, and nodes are connected by relationship edges.

Tools

Tool	Purpose
search_nodes	Search nodes by keyword, returning id, title, description, mastery
get_node	Get node details by node_id
create_node	Create a new node, and at the same time establish relationships via parent_node_ids / related_node_ids / contrast_node_ids Build relationships
update_node	Update node content (title, notes, tags, etc.)
update_mastery	Update mastery (delta range -1.0 ~ +1.0)

Mastery

Range 0.0 ~ 1.0, increased or decreased via delta. Adjust in small steps; do not make large changes at once.

Situation	delta
Answered correctly, clear understanding	+0.1 ~ +0.2
Answered correctly but hesitated	+0.05
Answered incorrectly, understood after explanation	0 ~ +0.05
Answered incorrectly and confused	-0.05 ~ -0.1
Explicitly stated they do not understand	-0.1 ~ -0.2

Creating Nodes

Node granularity (important)

One node = one independent concept. Criteria:

• ✅ "Limits", "Derivatives", "Binary Tree", "Quicksort" — single concepts, each independent
• ❌ "Limits and Derivatives", "Stacks and Queues", "TCP/UDP" — compound concepts, must be split

If the user mentions multiple concepts at once, create separate nodes and then connect them via relationships (e.g., related_node_ids）。

When to create

Only create nodes for concepts with independent learning value; do not create a node for every word mentioned.

Relationships (prevent isolated nodes)

Every new node should be connected to existing nodes in the graph.Must search thoroughly before creating, using the following strategies to find associations:

1. Direct search: search by concept name, e.g. search_nodes("导数")
2. Parent concept search: search the discipline/field it belongs to, e.g. search_nodes("微积分") or search_nodes("高等数学")
3. Related concept search: search for similar or prerequisite concepts, e.g. search_nodes("极限")(prerequisite knowledge for derivatives)

Try at least 2 or more search terms. If related nodes are found, they must be connected via relationship parameters:

Relationship type	Parameter	Scenario
Prerequisite prerequisite	parent_node_ids	Sub-concept / foundation that must be mastered first
Related related	related_node_ids	Same category / horizontal association
Contrast contrasts	contrast_node_ids	Opposite / contrast

Edge case: If there really are no related nodes in the graph (e.g., the user's first knowledge domain), first create a discipline/field-level parent node (e.g., "Advanced Mathematics", "Data Structures"), then create specific concept nodes under it.

Workflow

1. User asks a question → search_nodes find related nodes
2. Decide explanation depth based on mastery (low → basics, high → in-depth or quiz)
3. New concepts must be created in the current turn: search related nodes → create_node establish relationships → then explain. Do not wait until after explaining or until the next turn to create them.
4. User performance changes → update_mastery