Product blogs
Articles on positioning, architecture, and how the pieces connect.
Connect → Capture → Deliver — how people actually use ASB
A guided tour of the ASB map: five ways to connect, five peer capture/author methods, three deliverables — and zero Cloud/AI calls at runtime.
You can click it. You just can't teach it.
Self-drawn apps (SOUI and similar) accept clicks but hide UI names — why recording fails, and why plain-language specs plus screenshots are the right path.
Call local ASB skills from Cursor via MCP
Concrete recipe: enable ASB MCP, point Cursor at the local SSE endpoint, verify tools, and keep skill execution on your machine.
ASB vs classic RPA vs AI desktop agents
Three-way decision guide — when ASB fits, and when you should pick RPA or an AI agent product instead.
Security & data boundary (local-first)
What stays on-device at runtime, what the website handles, MCP exposure checklist, and what operators still own.
Spec-Driven: the one throughline across AI, harnesses, RPA, and ASB
Once a spec is refined to coordinates, waits, and parameters, the execution platform matters less — how AI, RPA, and ASB share one portable middle layer.
From 1–2 weeks to 1 day: automating a 230-step enterprise approval workflow
Real enterprise reimbursement approval — login, API risk flags, 4-branch logic — recorded in 90 minutes and packaged into 10 executables. 1 day delivery vs 1–2 weeks traditional RPA.
Video guides
Short Bilibili walkthroughs with links to full product blog articles (EN / 中文).
Three authoring methods — measured time and ROI
Compare pure AI code, modular record+exe, and record+refactor — with GLM-5 cost on OpenRouter and SiliconFlow and nesting ROI calculators.
From linear recording to loops and branches
Practical playbook: MCP-assisted recording, AI refactor to loops/if-else, and the three debug bottlenecks we measured.
From dynamic Excel to repeatable line entry
Splitting loop vs out-of-loop work, rewriting the operator prompt, and packaging one bill line as a reusable skill.
Your automation skills everywhere: MCP, VS Code, and mobile
Local-first MCP as a cross-platform agent hub — verified clients, ChatMCP and Tailscale on Android, and practical setup.
How to set up VS Code to leverage OpenRouter models
GitHub Copilot BYOK in VS Code: OpenRouter key, Chat: Manage Language Models, and catalog picks.
3rd party MCP usage and integration for the Automation Skill Builder
Product-first MCP roadmap, plus two patterns to wire third-party MCP with VS Code or Cursor today.
Product positioning vs. VM-first desktop AI
Where we sit when Hyper-V or mandatory sandboxes are not an option.
Ecosystem: development, broker, and runtime
How IDEs and MCP participate only during skill creation.
Works with popular AI services
OpenAI-compatible APIs, example providers, and model IDs—configure in Settings → Model selection.