Connecting an AI doesn't say what it's given without being asked: ask for consent first
Maintainers usually reply within 2 days
Nobody has claimed this yet.
Assessment
- Difficulty
- 5/5
- Estimated time
- Over a week
- Newbie friendliness
- 35/100
Research direction
Start by reading the AI connection and settings flows, then trace the history, memory_search, and memory_read tools and the background calls described in the issue. Map each passive source to where it is collected or sent, and determine how consent choices affect those paths. Done means consent appears before data is sent, each passive source has a default-off switch, the three requested choices work, and users can revisit the switches in Settings › AI.
Written by the indexing model from the issue text.
Description
The problem
Turning AI on in Yab and connecting Claude or Codex doesn't say what the AI will be given without being asked. I'd treat this as critical. It's the one moment a person decides whether to trust Yab with an AI, and today they decide without knowing.
I'm not talking about a page I point the AI at. If I ask it to read a page, of course it reads that page. The problem is the passive part: what Yab gathers on its own while I browse, and what a connected AI can reach or is sent without my asking. From reading the source (main, 0.47.19):
- Browsing history from every space, recorded as I browse: page titles, addresses and an index of page text. Any chat can pull it through the
historytool, whatever I asked. It includes history brought in from other browsers on the Mac (Arc, Chrome) when that switch is on. - Day threads and memory notes, written on their own: each day's "What you did" lines in
memory/episodic, whichmemory_searchandmemory_readhand to the AI. - Background calls to the model nobody asked for: with AI On, Yab sends page titles and opening text to the connected model to name the day's threads, and also makes link previews, form drafts and memory updates (including a daily pass over recent days).
- The tab in front, attached when a chat opens, and the list of open tabs in every space.
The only sentence that describes any of this is on the optional Know Me step ("Reads your browsing history and the sites you're signed in to."). Connecting the AI itself says nothing. I found out when I asked Yab's chat a plain factual question and it brought page titles from my own browsing into the answer.
What you'd like
A consent screen when AI is turned on or a new AI is connected, before anything is sent, that separates the two:
- When you ask: the AI sees the pages and tabs you point it at. (Expected, one line is enough.)
- On its own, unless you say no: each passive source above in plain words: history from every space and from other browsers, day threads and memory notes, background naming, previews and drafts, and the front tab on opening a chat. Each one gets its own switch, and the default is off until the person turns it on.
- Where it goes: to the provider the person connected, under that provider's terms.
- A real choice: "Allow all", "Only what I point it at" (passive sources off, history and memory tools unavailable) and "Not now".
Ask again when a new passive source is added, and keep the same switches in Settings › AI, so they can be changed later. Per-space privacy (#986) would sit alongside them.
How you manage today
I read the source to find out what was reachable, then denied the history and memory tools in my own Claude Code settings.
Related
#986 (per-space privacy), #871 and #872 (being clear about what leaves the Mac).
Drafted with Claude Code from my notes; the ask is mine.
- Dominant language
- Swift
- Stars
- 648
- Forks
- 56
- Avg merge
- 21h 14m
- Merged PRs (30d)
- 22
Getting set up
- No Dockerfile or Docker Compose file
- Has a pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from yetanotherbrowser/yab
-
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
yetanotherbrowser/yab#1008 ·
Maintainers usually reply within 2 days
-
area: passwords Broken bug
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
yetanotherbrowser/yab#923 ·
Maintainers usually reply within 2 days
-
bug needs triage
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
yetanotherbrowser/yab#918 · 1 reaction ·
Maintainers usually reply within 2 days
-
iPhone: the + button and the checkmark sit on top of the Space switcher in the bottom barPossibly taken @ChaseRoofener claimed this 3 days ago. OpenAnnoying area: iphone bug needs info
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
yetanotherbrowser/yab#915 · 3 comments ·
Maintainers usually reply within 2 days
-
CONTRIBUTING.md and CI run scripts/runs/tests_for.py, which is not in the repositoryPossibly taken @serafimcloud claimed this 3 days ago. OpenAnnoying in progress
Difficulty 2/5 1-3 hours Newbie friendliness 74/100
yetanotherbrowser/yab#875 · 1 comment ·
Maintainers usually reply within 2 days
All issues in yetanotherbrowser/yab
Similar issues
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 70/100
swiftlang/swift-org-website#1608 ·
Maintainers usually reply within 2 days
-
Difficulty 2/5 1-3 hours Newbie friendliness 65/100
haacked/switchboard#460 ·
Maintainers usually reply within 1 day
-
Difficulty 1/5 1-3 hours Newbie friendliness 78/100
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 67/100
manaflow-ai/cmux#47572 ·
Maintainers usually reply within 1 day
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
borrowcheckmate/swimlie#48 ·
Maintainers usually reply within 1 day