How Does Automatic Entity Extraction Work?
When you save a memory, SiderMem analyzes its content to identify the entities and topics it touches on, then matches them against your existing knowledge base — new topics become new wiki pages, and anything overlapping an existing topic is merged into it, with no manual tagging involved.
The pipeline, step by step
- You save a memory — a web page, note, file, or AI chat.
- Once it syncs to the cloud, extraction reads the memory and returns a list of entities and topics it touches on (not a single category).
- Each entity is resolved against your existing wiki pages — a new topic creates a new page; an existing topic gets this memory added to its sources.
- Only the pages actually touched by this memory are regenerated, aggregating everything ever saved about that topic.
Why one memory can touch several pages
Extraction doesn't force a memory into one bucket. A memory discussing a project and a specific tool used on it can update both the project's page and the tool's page — this is what gives SiderMem automatic cross-card entity merging, and it's the same underlying structure that Graph View visualizes as connected nodes.
What it costs
Entity extraction is included with cloud sync capacity and is never metered against your AI Summarization or Ask quota — it reads your original saved content, not an AI-generated summary, so there's no overlap with the credit-consuming summarize feature.
Limitations
Extraction only runs on synced memories, so local-only memories won't appear in your wiki until you sync them. Very short or ambiguous content may extract fewer or broader entities than a longer, specific memory would.