DocWatch, watched folders
Files enter your memory through one door: the watched folder. Point a vault at a folder, and whatever lands there gets read, cut into chronicles and vectorized, automatically, and again whenever it changes.
How fast? Watch a real wave land, 92 files vectorized in about fifteen seconds, a researcher's corpus flowing into their Recherche vault, live:
Reading the bar
That thin strip at the top of the app is DocWatch talking:
- Indexing ×N: the counter of files taken in this wave.
- Vectorizing: each file is being cut into chronicles and embedded. The
768Dbadge is how many numbers each memory's fingerprint holds, which is set by the model you chose in Models & credits. - Memory consolidation: the wave settles into the vault's indices.
And in the System Diagnostics widget,
DocWatch Status · Active confirms the watcher is on duty.
What it reads
Anything else is refused, never guessed. A vault only accepts what it can actually read.
Governed, not automatic
DocWatch is fast precisely because the governance around it is strict:
- A standing consent, per folder. Auto-ingestion is a policy YOU grant to a specific path, never a global switch, never all-auto by default.
- Big waves get a gate. When a large batch of new files appears at once, the OS asks before ingesting, so a mistaken drop of a thousand files never storms your memory.
- New arrivals mature. Fresh files wait in quarantine before joining your memory's answers, so a wrong file never contaminates instantly.
- Traceable and revocable. Every policy is listed, every ingestion attributable to its folder, and revoking a consent is one click.
Where to set it up
Every vault's Sources section holds its watched folders. Add one, rescan, or detach it anytime. The walkthrough lives in Your first vault, and web capture uses this same door: everything you capture lands as a file in a watched folder, and DocWatch does the rest.
From 1.4.3 the same door takes the memory folder of a coding agent: connect your coding agent's memory.