DEV Community

Fenju Fu
Fenju Fu

Posted on

Agents Have Safe Runtimes and Multi-Engine Orchestration — But Who Operates the Desktop?

#aiagents #rpa #automation #opensource

Today's GitHub Trending tells a clear story: Agent infrastructure is maturing rapidly.

  • NVIDIA/OpenShell — "The safe, private runtime for autonomous AI agents." A sandboxed execution environment, backed by NVIDIA, built in Rust.
  • mvschwarz/openrig — "Multi-agent harness that runs Claude Code and Codex together as one system."

These are fantastic contributions. The runtime layer (how agents safely execute code) and the orchestration layer (how multiple agents coordinate) are getting solid open-source solutions.

But here's the question every developer actually hits in practice:

When your agent needs to fill a form in an internal OA system, scrape a report from a browser dashboard, or click through a legacy desktop app — what does it do?

In most cases, it writes a Python script and tells you to run it yourself.

The Missing Layer: Desktop Execution

We now have:

  • ✅ Safe execution runtimes (OpenShell)
  • ✅ Multi-engine orchestration (openrig)
  • ✅ Context optimization (context-mode)
  • ❌ Desktop and browser automation for agents

This is the gap that iflytek/astron-rpa fills.

What astron-rpa Does

astron-rpa is an Agent-ready RPA suite. Instead of asking agents to write scripts for you to run manually, it lets agents directly operate desktop software and browsers — performing unattended automation of real UI workflows.

Think of it this way:

Layer What it does Example project
Safe runtime Sandboxed code execution NVIDIA/OpenShell
Orchestration Multi-engine coordination mvschwarz/openrig
Desktop execution UI automation, browser ops, form filling iflytek/astron-rpa

Astron RPA desktop automation interface

Pairing with Workflow Orchestration

Desktop execution alone isn't enough — you need a workflow platform to orchestrate the full pipeline: trigger condition → RPA execution → data processing → result writeback.

That's where iflytek/astron-agent comes in. It's an enterprise-grade agentic workflow platform that can orchestrate multi-step business processes, combining RPA execution with data processing and decision-making.

Astron Agent workflow canvas

The Full Stack

The emerging Agent stack looks like this:

  1. Runtime: Safe sandboxed execution (OpenShell-style)
  2. Orchestration: Multi-agent coordination (openrig / astron-agent)
  3. Execution: Desktop and browser automation (astron-rpa)

Layers 1 and 2 are getting well-deserved attention. Layer 3 is where iflytek/astron-rpa comes in.

👉 Explore: https://github.com/iflytek/astron-rpa
👉 Orchestrate: https://github.com/iflytek/astron-agent

Top comments (0)