Update (11 Nov 2026): ScrollRSI is now on Google Play: https://play.google.com/store/apps/details?id=app.scrollrsi&referrer=utm_source%3Ddevto%26utm_medium%3Dorganic%26utm_campaign%3Dlaunch%26utm_content%3DA7b. US$2.99 one-time, Android 11+.
I build with several AI agents at once. The deep wiring between them lives in code, but there is always a hands-on layer left: I read what one agent produced on my phone, and the next agent needs to see all of it.
The annoying part
On a phone, that hand-off used to mean select, copy, scroll, repeat, one chunk at a time, across a chat that might have run all day. Most chat apps don't give you an export button for a thread. And a half-copied context is worse than none, because the next agent confidently works from the half it got.
What I do now: one-tap thread sharing between apps
I made ScrollRSI, a small Android app, for exactly this layer. One tap starts it; after that it's hands-off:
- Tap the floating bubble inside the agent app I'm reading. No switching apps to start it.
- It auto-scrolls the whole thread while recording, even a very long one, and stops at the end by itself (it uses Accessibility only for the swipes and screenshots of a capture I start).
- It transcribes on the phone. An open-source OCR engine runs on-device, drops the repeated lines from overlapping screens and stitches them in order. The app has no internet permission, so nothing is uploaded.
- Then I share the file. I get one clean, ordered .txt, named after the page, in Download/ScrollRSI with the recording saved beside it, and attach it to the next agent: Grok, Gemini, ChatGPT, Claude, or anything that takes an attachment.
Why this matters for agent context
The part I care about most as an agent user: the file flags what might be lost. If a link was cut off or text was hidden behind a reply box, the file says so instead of guessing, and every file ends with a run footer so you can check the result. So the next agent isn't silently working from a gap. There's also an optional clean copy that tidies layout only; words and digits stay exactly as they were on screen.
Where it fits (and where it doesn't)
This doesn't replace a harness or an API pipeline. It's for the moments you are the router: testing a prompt chain by hand, getting a second model to review the first one's reasoning, or picking up yesterday's long session in a different assistant.
Details
Android 11+, US$2.99 one-time, no accounts, no ads. It's on Google Play.
More at https://scrollrsi.kubeai.info
If you can scroll it, you can save it. How are you moving context between agents on mobile today?
Top comments (0)