APEMIND / KNOWLEDGE GRAPH
Follow connections. Find the source.
Explore people, products and concepts across your documents. Use the graph to find a reading path, then inspect the passages behind each connection.
A knowledge graph for investigating your documents
ApeMind combines graph, vector and full-text retrieval. Start with an entity, explore its neighbors and return to the source material. Agents can use graph tools alongside document reading to investigate a question.
Review entity merge and noise-removal suggestions with their sources. Confirmed governance rules have an operation history and can be reverted. A graph connection helps locate evidence; the original passage determines what it means.
Development preview
Topic discovery in Knowledge Graph v2
Topic maps and the entity-card workflow below are in development. Availability depends on your installed version and collection; they are not included in every deployment.
Discover topics
Community detection groups connected entities into topics and selects representative entities. Start with a topic when you are not yet sure what to search for.
Follow a connection
Entities shared across topics suggest a path worth investigating. Open one to see where it appears and which other entities are mentioned with it.
Check the passage
Inspect the quoted text and open its source document. Use entity lookup tools to let an agent investigate the same material and continue reading.
The preview uses weighted Louvain community detection on entity co-mentions. Smaller or less varied collections may show an entity map instead of topic bubbles. A shared mention is not, by itself, a semantic relationship or a causal claim.
A reading path through an unfamiliar library
Suppose a research collection covers energy storage, materials and manufacturing. A material appearing in two topics can lead you to documents on both. Read those passages before concluding that a manufacturing process uses that material. This is an illustrative workflow, not a measured customer result.
When is a graph worth evaluating?
Use entities to explore
Try a graph when you need to investigate a named object across documents or inspect connections among several objects.
Keep direct retrieval available
Full-text and vector search remain useful for exact phrases, direct facts and passages that entity recognition missed.
Compare on your own questions
Hold documents, model and tool budget constant. Check answer support, omissions, latency and model usage; count graph-building costs separately.
Questions about knowledge graphs and GraphRAG
What does community detection add?
It provides a way to browse groups of connected entities. In the v2 preview, topic representatives and cross-topic entities help choose where to read next. The groups are not a complete classification of every document.
Is a topic map the same as a whole-library summary?
A topic map organizes exploration. Short topic labels and suggested questions do not establish a complete community-report or global question-answering pipeline.
Does the graph make every answer more accurate?
There is no universal improvement guarantee. Recognition can miss entities, and shared mentions can be misleading. Verify the original passages and compare results on representative questions.
Can agents use the graph?
Agents can combine graph queries with source reading. The v2 preview adds entity cards and related-entity lists with quotations. Tool availability still depends on the collection, permissions and enabled capabilities.
Learn how to explore and verify
The detailed guides below are in Chinese.
Start with a question you can verify
Use a collection you are authorized to access. Find an entity, inspect the sources and compare the result with ordinary retrieval. Preview-only controls may not be available in your version.
Open your knowledge bases