Memory Datasets
A memory dataset is a container for knowledge — think of it as the library an assistant reads from. It sits at the center of everything in Seekdown: content flows into a dataset, and assistants (and integrations) read out of it. Nothing an assistant says comes from anywhere else, which is why the quality of an assistant follows, almost one-to-one, the quality of its dataset.
For example, a support team keeps a single dataset that combines their public help center (captured by a data capture job) with a handful of internal policy PDFs (uploaded by hand). One public assistant on their website and one internal assistant for staff both read that same dataset — same knowledge, two audiences. When the help center changes, they re-crawl once and both assistants are up to date.
What a Dataset Holds
- Website content captured by a data capture job.
- Uploaded documents — PDF, Office files, text, and more.
- Content pushed in through the content API (Enterprise plan).
- Form submissions (leads) collected by an assistant.
Whatever the source, every item is processed and indexed so it can be searched. That processing is why a freshly added file isn't answerable the instant it lands — it becomes searchable a moment later, once indexing finishes.
What You Can Do With One
Add and edit content, import files in bulk, search and filter what's inside, export the whole thing (Excel, CSV, or JSON), generate keys for integrations, and clean it (empty it but keep the container) or delete it. Crucially, a single dataset can feed several assistants and be filled by several data capture jobs at the same time — it's a shared hub, not a private silo.