0.85.0
takahirom/arbigent0.85.0Sep 17, 2026by takahirom
Upgrading?
AI Summary
This release improves the pacing logic for replayed steps, specifically for bare key presses. Instead of sleeping out the full recording interval, the tool now waits for the device's focused element to match the recorded target before proceeding, resulting in faster replay times while maintaining the original timeout ceiling.
Key Highlights
- Replayed steps with no target now wait for the recorded focus element to match the device's current focus.
- Performance is significantly improved for steps driven by D-pad inputs on Android TV.
- The original ceiling timeout is still respected if the recorded focus cannot be found.
- Traces recorded before this version continue to pace exactly as they did.
New Features
- Smart focus waiting for replayed steps without explicit targets
- Dynamic pacing based on UI focus state instead of fixed sleep intervals
Full Release Notes
A replayed step with a recorded target waits for that target. A bare key press has none — which is most of a run driven by a D-pad, as on Android TV — so replay slept out the interval the recording happened to have. ## Replay Paces on Focus (#448) A replayed step with a recorded target waits for that target. A bare key press had none, so replay slept out the interval the recording happened to have. Those steps now wait until the device's focused element matches the one recorded for the step: ``` Replay wait: focus android.view.ViewGroup at [107,520,277,617] found after 142ms before capturing step 1 ``` The recorded gap stays the ceiling, so a step that never reaches its recorded focus waits as long as it did before. On one emulator and one scenario of 12 replayed steps, the same traces took 154s before and 75s after. When focus cannot tell the screen before the action from the screen after it — same position recorded twice, no predecessor, initializers, or a previous task that fell back — the wait first has to observe a screen that is not the recorded focus. Traces recorded before this release pace exactly as they did. ## What's Changed * Wait for the recorded focus on replayed steps with no target by @takahirom in https://github.com/takahirom/arbigent/pull/448 **Full Changelog**: https://github.com/takahirom/arbigent/compare/0.84.0...0.85.0
Related Repositories
BreezeAPP 是一款為 Android 和 iOS 平台開發的純手機 AI 應用程式。從 App Store下載,即可在不連網的狀態下享受多項 AI 功能。源碼由聯發創新基地(MediaTek Research)提供。我們旨在推廣兩個概念: 人人都可以在自己的手機上自由選擇並運行不同的LLM - one is free to choose one's own LLM to run on a phone,以及任何app開發者都可以輕鬆寫作創意的純手機AI應用 - any dev can create purely phone-based AI apps easily。
A collective list of free APIs
The AI that really does things. Any OS. Any Platform. The lobster way. 🦞
Interactive roadmaps, guides and other educational content to help developers grow in their careers.
Reverse engineer anything with agents, from app behavior down to native binaries.