0.87.0
microsoft/markitdown0.87.0Sep 25, 2026by takahirom
Upgrading?
AI Summary
This release drastically reduces the number of view hierarchy fetches per step, cutting non-AI execution time. It also improves error handling by reconnecting on failure rather than probing before every read.
Key Highlights
- Hierarchy fetches reduced from 11-13 to 3-4 per step
- Reconnects on failed device read
- Elements and focus come from a single hierarchy fetch
New Features
- Reduced hierarchy polling overhead
- Improved connection resilience
Full Release Notes
Each step now reads the device screen with far fewer view hierarchy fetches, which cuts non-AI time per step. On an Android emulator, reading the screen went from about 1.4s to 0.2–0.4s per step, and hierarchy fetches went from 11–13 to 3–4 per step. ## Fewer Hierarchy Fetches per Step (#452, #453) - Reads no longer check the connection with an extra hierarchy fetch first. A read that fails reconnects and is retried. Actions keep the up-front check, so a tap or key press that failed part way is not repeated blindly. - A step's elements, UI tree and focused element now come from one hierarchy fetch, so they always describe the same frame. ## What's Changed * Reconnect on a failed device read instead of probing before every read by @takahirom in https://github.com/takahirom/arbigent/pull/452 * Read a step's elements, tree and focus from one hierarchy fetch by @takahirom in https://github.com/takahirom/arbigent/pull/453 **Full Changelog**: https://github.com/takahirom/arbigent/compare/0.86.0...0.87.0
Related Repositories
public-apis/public-apis★ 491,123
A collective list of free APIs
NousResearch/hermes-agent★ 255,883
The agent that grows with you
yt-dlp/yt-dlp★ 199,986
A feature-rich command-line audio/video downloader
Significant-Gravitas/AutoGPT★ 187,698
AutoGPT is the vision of accessible AI for everyone, to use and to build on. Our mission is to provide the tools, so that you can focus on what matters.
langgenius/dify★ 159,154
Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.