experiments in a post-browser web
fix(izui): promote content to child-content when opened from parent window
Extensions explicitly passing role:'content' bypassed the inference that
detects parentWindowId and assigns child-content. This meant ESC did
nothing for pages opened from groups (escPolicy returns 'nothing' for
content role in active session).
Fix: after role assignment, promote content → child-content when
parentWindowId is set. Background/extension-host windows have null
parentWindowId so they're unaffected.
+27
-2
+10
backend/electron/ipc.ts
+10
backend/electron/ipc.ts
···
1861
1861
role = 'utility';
1862
1862
}
1863
1863
}
1864
+
1865
+
// Phase 2: Promote 'content' to 'child-content' when opened from a real parent.
1866
+
// Extensions may explicitly pass role:'content' for web pages, but if the IPC
1867
+
// sender is a visible window (not background/extension-host), the opened page
1868
+
// is semantically a child — ESC should close it and return to the parent.
1869
+
if (role === 'content' && parentWindowId != null) {
1870
+
console.log(`[izui] Promoting content -> child-content (parent: ${parentWindowId})`);
1871
+
role = 'child-content';
1872
+
}
1873
+
1864
1874
console.log('[izui] Window role:', role, 'for:', url);
1865
1875
1866
1876
const windowParams = {
+17
-2
peek-todo.md
+17
-2
peek-todo.md
···
243
243
Browsers do not not acknowledge entities across pages.
244
244
If you load two pages about the band Tortoise, the browser only knows you loaded two different domains.
245
245
Peek's entity-centrism projects uplifts the apex of understanding from publishers, into the user agent.
246
+
246
247
The application identifies and records entities in the pages you load.
247
248
Aims for approximate equivalancy via confidence threshold scoring.
248
249
The expirement starts with coarse types, eg person, place, event, organization.
···
255
256
Uses the sourcing data to infer links between entities.
256
257
Scores entities based on frecency of observation and interaction.
257
258
258
-
- [ ] get people, places, dates/times/events
259
-
- [ ] get meaningful numbers, and their label, eg phone
259
+
Entity detection first take is using regexes against page content.
260
+
Can experiment also with https://github.com/mozilla/readability which we'll need for other features as well.
261
+
Can harvest microformats with https://github.com/microformats/microformats-parser, index as the relevant entity type.
262
+
We'll experiment with offline NER machine learning models as well, including training micro-models here from the data we're collecting.
263
+
264
+
We'll cast a wide net to start, recording everything.
265
+
Over time, we'll learn more about our needs and the limits of this approach.
266
+
Then we can design expiration policies, archiving, import/export, etc.
267
+
But the default should be to never lose anything.
268
+
269
+
Initial entities to harvest:
270
+
271
+
- [ ] people and related info, eg homepage, nick, social, email
272
+
- [ ] places
273
+
- [ ] dates/times/events
274
+
- [ ] meaningful numbers and their label, eg phone numbers, tracking numbers, booking numbers
260
275
261
276
- [ ] extract a table as csv
262
277
- [ ] layer outside of web page, and in between pages (eg event page -> event -> any calendar page)