Memory Graph¶

Link related memories to build a navigable knowledge graph. The graph enables relationship-aware retrieval and reasoning.

Creating Relationships¶

# Create typed relationships
mem.relate("/events/2026-03-15/bug",     "/knowledge/auth/jwt-expiry",  "caused_by",  strength=0.92)
mem.relate("/knowledge/auth/jwt-expiry", "/knowledge/auth/architecture", "related_to", strength=0.85)
mem.relate("/events/2026-03-14/deploy",  "/events/2026-03-15/bug",      "follows",    strength=1.0)

llmfs relate /events/2026-03-15/bug /knowledge/auth/jwt-expiry caused_by
llmfs relate /knowledge/auth/jwt-expiry /knowledge/auth/architecture related_to --strength 0.95

Relationship Types¶

Type	Meaning
`related_to`	Generic semantic connection
`follows`	Temporal succession (A happened after B)
`caused_by`	Causal link (A was caused by B)
`contradicts`	Conflicting information

Graph Traversal¶

Via MQL¶

SELECT memory FROM /knowledge RELATED TO "/events/2026-03-15/bug" WITHIN 2

Via Python API¶

from llmfs.graph.memory_graph import MemoryGraph

graph = MemoryGraph(mem._db)
neighbors = graph.get_neighbors("/knowledge/auth/jwt-expiry")
path      = graph.traverse("/events/2026-03-15/bug", max_depth=3)

Traversal supports both BFS (breadth-first) and DFS (depth-first) with configurable depth limits.

Auto-Linking¶

LLMFS can automatically link semantically similar memories on every write():

mem = MemoryFS(auto_link=True, auto_link_threshold=0.75, auto_link_k=3)

After writing a memory, the top-k most similar existing memories are linked with related_to edges if their similarity exceeds the threshold. Disable with auto_link=False.

How It Works¶

Graph edges are stored in the relationships table in SQLite:

source_id / target_id: Memory file IDs
type: One of the four relationship types
strength: Float from 0.0 to 1.0
created_at: Timestamp

Traversal results include visited paths, edges, and a depth map showing how far each memory is from the starting node.