Iwtbs27/collab-agent

What it solves

OmniAct addresses the limitation of single-environment agents (such as those restricted only to coding, browsing, or desktop apps) by providing a unified framework that allows a single agent to complete complex digital tasks that span across multiple environments, including web browsers, desktop applications, and the command line (CLI).

How it works

The system uses a hierarchical architecture centered around a "TeamLeader" and specialized "Workers":

• TeamLeader: Acts as the orchestrator that analyzes the user's goal and delegates specific sub-tasks to the appropriate worker.
• Specialized Workers: Dedicated agents for the Browser, CLI, and Desktop environments. Each worker maintains its own observation and action loop tailored to its specific environment.
• Task Directory: Collaboration is managed through a structured file system (tasks/<task-id>/). Workers communicate by writing requests, results, and artifacts (like screenshots or files) into this directory rather than sharing internal state.
• Human-in-the-loop: The system supports manual takeover, allowing users to intervene, resume, or cancel specific worker tasks.

Who it’s for

Users who need to automate complex workflows that require switching between different software environments on a Windows machine, such as moving data from a web browser to a desktop app or executing CLI commands to process files.

Highlights

• Cross-environment execution: Seamlessly transitions between Browser, CLI, and Desktop workers.
• Transparent execution: Maintains a detailed execution trajectory and task directory for auditing and debugging.
• Artifact-based handoff: Workers deliver concrete results (Markdown summaries and files) instead of simple completion messages.
• Session persistence: Supports inheriting Edge browser profiles to maintain login states in isolated worker environments.