---

name: "advanced-graphs-pipelines" description: "Guides Cognee custom graph schemas, ontologies, custom pipelines, memify, migration, export, and visualization workflows." disable-model-invocation: true metadata: disco-role: operating license: Apache 2.0

Cognee Advanced Graphs and Pipelines

Use this sub-skill when the user is beyond the default memory flow and wants to shape graph extraction, run custom tasks, enrich a graph, migrate memory, export graph data, or visualize the result.

Route here for

• Custom DataPoint models and graph extraction schemas.
• Dedup, Embeddable, and LLMContext field annotations.
• Custom Task / @task / run_custom_pipeline(...) workflows.
• cognify(..., graph_model=..., custom_prompt=..., temporal_cognify=...).
• memify(...), ontology resolver configuration, migration sources, export, and visualization APIs.
• Troubleshooting model identity, invalid schemas, custom-task failures, and visualization/export issues.

Route away

• Basic memory storage/query flows: core-memory.
• Search-mode selection: search-retrieval.
• Provider/database configuration: configuration-backends.
• Session/agent memory and feedback: agent-session-memory.
• CLI/API/MCP service startup: api-cli-services.

Read first

1. references/custom-graphs-pipelines.md
2. references/migration-visualization.md
3. references/troubleshooting.md

Safe helper

• scripts/inspect_custom_model.py — imports a custom model and prints Pydantic fields/metadata without running a pipeline.

Working rules

• Import DataPoint from cognee.infrastructure.engine, not from internal module paths.
• Use identity fields for domain objects that must deduplicate across runs.
• Keep custom pipeline tasks small, async-safe, and explicit about whether they call LLM/embedding providers.
• Use skip_connection_test=True only when the custom pipeline truly performs no LLM or embedding work.
• Treat migration/export/visualization as graph operations that require an already initialized dataset and backend.