Context compaction happens in the dark. I made it happen on a map.

ThoughtDAG presents AI conversations as a graph on a canvas and makes compaction transparent.

Instead of silently rewriting history, it finds compressible runs of turns and lets users review or exclude them.

The compressed result is saved as a separate tree, while marked decisions and highlights are preserved.

Each distilled node links back to its source, and users can compare full and condensed outputs to see what changed.

The approach turns context-window compression into an inspectable workflow that reduces hidden information loss in long AI sessions.