Turn notes into a memory structure you can inspect.

Build a transparent L0 to L3 map from local text. Every derived record keeps its source IDs, so grouping stays reviewable instead of disappearing into an opaque embedding.

Source notes

one note per line

Conversation notes

The sample is illustrative. It does not call an agent, model, or database.

Ready for notes.

Interpretation boundary. This local tool deterministically groups parsed notes. It does not implement TencentDB Agent Memory, SQLite persistence, semantic inference, token reduction, or live agent context.

L0 conversation0

L1 atom0

L2 scenario0

L3 persona0

Graph edges0

Traceable records

No map built

Your completed hierarchy will appear here.

Persisted locally after each completed build.

Relationship graph

Mermaid-style text

Build a pyramid to generate graph text.

Trace the source note-to-pyramid record rules

Your browser does not support video.

Read the explanation

The written builder parses sentences, then creates one conversation record and one normalized atom for each parsed sentence. Four sample sentences therefore make four L zero and four L one records. These bars use one hundred twenty pixels per record. Atom source IDs point back to conversation IDs. This is deterministic local text processing, not a semantic memory model or agent inference. All four sample atoms point to one scenario, and that scenario points to one persona record. Four conversations plus four atoms plus two upper records total ten nodes. The graph has four preserves edges, four groups edges and one summarizes edge, total nine. Both bars use fifty pixels per graph item. The grouping is fixed to one scenario, not inferred topic clustering. For N parsed notes, total nodes are two times N plus two, and edges two times N plus one. A one-sentence custom note set makes four nodes; two sentences make six. Bars use eighty pixels per node. IDs such as L zero one are scoped to this build rather than durable cross-build identities. The completed JSON is stored in browser local storage; no SQLite, remote service or model call runs.