Skip to main content
Canvas is a visual answer to “how does this connect to the rest of my work?” It renders your files as a constellation on the GPU, laid out by the relationships FilDOS already knows about.
The Canvas knowledge graph

​
Three kinds of connection

Canvas fuses three signals into one graph. Toggle each with the chips at the top:
  • Similar — files whose embeddings are close in meaning (per-file centroids of their stored chunk vectors, compared with a partitioned k-nearest-neighbour search).
  • Entities — people, places, and organizations pulled from your documents by an on-device NER model, shown as mint diamonds.
  • Tags — your colour-coded tags become star nodes.
  • Sessions — files you worked on around the same time, derived from modification times.
Nodes are grouped into communities (Louvain clustering) and painted with the brand’s six accent colours, so related clusters read at a glance.

​
Exploring the graph

  • Find in graph — search to spotlight matching nodes.
  • Detail panel — click a node to see why it’s connected to its neighbours.
  • Time scrubber — an mtime histogram at the bottom lets you replay how your files accumulated over time.

​
Built lazily, kept fresh

Canvas does no work until you open it. After that it’s incremental: only changed files are recomputed, and the whole build runs under the same duty-cycle as the indexer, so it never fights your foreground work.
The NER model that powers Entities is opt-in and never auto-downloaded — Canvas works without it. Enable it and rebuild from Settings → AI & Search → Canvas.