Oasis Browser: Technical and Interaction Data
One thing we care a lot about with Oasis is being explicit about what leaves the browser when you use the assistant. Not a vague "privacy-first" label, but the actual JSON.
Insight for other users: By default, Oasis does not send your email, user ID, or a user block to Kahana. Interaction payloads are anonymized: things like your prompt/response text, token counts, latency, session ID, browser/OS version, and basic context (for example active tab URL and title for that interaction). That is it unless you opt in under Settings → Privacy (about:preferences#privacy), where you can turn on personalization and include account fields (email, user_id, locale, opt_in_data_collection_use: true).
We also do not use this to build ad profiles. No location history for ads, no search or browsing history sold to brokers, no interest graphs for third-party advertising. The goal is product improvement: quality, latency, bugs, and aggregate feature usage. Not selling your data.
You can see side-by-side example payloads (default vs opted-in) here:
https://kahana.co/docs/technical-and-interaction-data
Questions we would love your take on as builders and users:
Default-off personalization
Is anonymized-by-default the right default for an AI browser, or would you want opt-in framed differently (for example clearer "train Oasis for me" vs "help improve Oasis for everyone")?What is in the payload
For the fields we do send by default (prompt/response text, active tab URL/title, session ID), is anything still too much for your comfort? What would you want removed or hashed?Transparency format
Does publishing real-shaped JSON examples (with an interactive toggle on the doc page) actually build trust, or would you prefer a shorter plain-English summary only?Firefox + AI
For people who want Firefox-level privacy and a built-in assistant: what is the minimum you need to see before you would switch from Chrome or a Chromium AI browser?For the Oasis team
If you have tried Oasis, did the Privacy setting match what you expected after reading the doc? Any mismatch between "what I thought was local" and "what left the browser"?
Happy to answer specifics in the thread. We are iterating quickly and your feedback directly shapes what we collect and how we document it.


Replies