Reflect

Reflect is the “thinking” operation. It analyzes the knowledge graph and improves it by organizing, synthesizing, and restructuring.

Data Flow

[1. Discover]        → find orphans, weak links
    │
    ▼
[2. Cluster]         → group by embedding similarity
    │
    ▼
[3. Gather Context]  → load existing Domains and Concepts
    │
    ▼
[4. Analyze (LLM)]   → decide per cluster
    │
    ▼
[5. Persist]         → apply changes to graph

Discover

Find memories that need attention: orphaned facts without parent concepts, weak or missing links between related memories.

Cluster

Group related memories by embedding similarity. Memories that are semantically close are placed in the same cluster.

Gather Context

Load existing Domains and Concepts so the LLM knows what organizational structure already exists.

Analyze

For each cluster, the LLM decides:

• Link facts to an existing concept
• Create a new concept or domain
• Synthesize new memories from patterns
• Restructure relationship edges

Persist

Apply all changes to the knowledge graph atomically.

Semantic Hierarchy

Reflect builds and maintains the layer hierarchy:

"Programming" (Domain)
  └── "Type Systems" (Concept)
        ├── "Haskell has ADTs" (Fact)
        └── "Rust has ownership" (Fact)

Synthesis

Reflect can draw inferences from existing knowledge:

• Fact A: “All mammals are warm-blooded”
• Fact B: “Whales are mammals”
• Synthesized: “Whales are warm-blooded” (inferred)

When to Use

• After bulk ingestion of many facts
• Periodic maintenance to organize accumulated knowledge
• When recall quality degrades due to unorganized memories

API

curl -X POST https://api.memlayer.dev/api/v1/reflect \
  -H "Content-Type: application/json" \
  -d '{"query": "programming", "dry_run": false}'

Field	Type	Required	Description
query	string	No	Only consolidate memories matching this topic
dry_run	bool	No	Preview operations without applying (default: false)
threshold	float	No	Minimum similarity for linking facts
cluster_threshold	float	No	Minimum similarity for clustering
limit	int	No	Maximum facts to process