Ctrl K

Context

Graphify for code, Cognee for everything else

Ask the graph

Cognee recall over the context layer

Knowledge graph layers

Loading Cognee graph…

Cognee layer

Context & memory

Telegram chat logs
Project plans & OKRs
Personal habits
Financial receipts
Research papers / PDFs

Graphify layer

Code & AST engine

C++ / Python repos
Next.js UI component trees (21st.dev)
API routes & schemas
Function call graphs

High signal-to-noise for Cognee

Codebases emit millions of granular AST nodes — variables, syntax tokens, imports. Dumping source into Cognee dilutes its vector-graph memory. Kept apart, Cognee only indexes what is being built and why.

Deterministic precision for Graphify

Vector search is probabilistic; it guesses by similarity. Code is deterministic. Tree-sitter static analysis maps exact imports, classes and dependencies, so SecOps and Editor get non-hallucinated traces.

Zero-overlap partitioning

Both graphs live in the same local Memgraph instance, separated by graph namespace — no cross-contamination, one database to run and back up.

Cognee context — Advisor / Tutor

MATCH (u:User)-[:HAS_GOAL]->(g:Goal)
      -[:REQUIRES_KNOWLEDGE]->(c:Concept)
RETURN g, c;

Graphify structure — Editor / Hacker

MATCH (f:Function {name: "execute_dax_query"})
      -[:CALLS]->(dep:Function)
RETURN f, dep;

Which layer each persona reads

Persona	Layer	Queries
Advisor	Cognee	Baseline, historical performance, 6-week milestones, calendar availability
Librarian	Cognee	Saved whitepapers, NotebookLM shelf metadata, filtered research notes
Tutor	Cognee	Weakness logs, active recall history, Socratic confusion patterns
Roommate	Cognee	Daily study hours, expense tracking, focus logs, energy levels
Editor	Graphify	Submission ASTs, function complexity, docstrings, variable scoping
SecOps / Hacker	Graphify	Hardcoded secrets, static vulnerability paths, dependency CVEs