auwsx: Project-Oriented Autonomous Workspace Orchestrator and Agent Scheduler System

auwsx is an autonomous workspace orchestrator implemented in Rust, using a separated architecture of daemon process and TUI client, providing continuous task scheduling, to-do management, and deterministic execution pipelines for coding agents. As a sister project of wsx, it focuses on headless autonomous workspace orchestration capabilities, aiming to solve the problem where AI agents require continuous human participation in traditional interactive modes, and supports long-term autonomous completion of complex coding tasks.

With the widespread application of LLMs in software development, developers' demand to explore AI agents that can continuously and autonomously complete complex coding tasks is growing. Traditional interactive modes require continuous human participation, while a truly autonomous workflow needs a system that can run long-term, schedule automatically, and manage complex task states. auwsx emerged as a sister project of wsx, focusing on providing headless autonomous workspace orchestration capabilities.

auwsx uses a layered architecture:

1. Data Layer: Persistent storage based on SQLite, covering entities such as projects, to-dos, issues, routine tasks, scheduling runs, agent runs, etc., ensuring complete state recovery after system restart.
2. Execution Layer: A long-running daemon process provides IPC communication via Unix Socket, using a single-point write strategy to avoid concurrency conflicts. Its responsibilities include automatically triggering project processing, managing to-do approval classification, executing issue state transitions, and coordinating agent subprocesses.
3. Interaction Layer: A TUI client based on Ratatui, using a two-column layout (left side: project tree view, right side: details, bottom: status bar). All data operations are done via daemon IPC to ensure architecture separation.

auwsx implements a deterministic issue processing pipeline: CONSOLIDATING → PLANNING → PLANNED → IMPLEMENTING → REVIEW → AUDIT / NEEDS_FIX / CONFLICTED → ENDED → COMPLETING → DONE. Each stage has clear entry conditions and exit rules. The scheduler only selects issues in actionable states for execution, and avoids duplicate scheduling through concurrency checks.

Scheduling Interval Configuration: Each project can independently set schedule_interval_min (None = manual trigger only; Some(0) = check every tick; Some(n) = at least n minutes interval). Human Intervention: Intervene via TUI shortcuts (E triggers scheduling/execution, a for approval, T for classifying to-dos). Approval Classification: To-dos require approval by default; those from human/inbox sources are already approved, while those from agent/routine sources are pending approval. After approval, they can be classified into specific issues.

auwsx creates an independent working directory for each issue to ensure code change isolation, no contamination of the main branch, and independent reset of failed/conflicted issues. Default storage paths: ~/.local/share/auwsx/state.db (core database), ~/.local/share/auwsx/runs/ (agent logs), ~/.local/share/auwsx/main-jobs/ (main job records), ~/.auwsx/inbox/ (inbox). The default paths can be overridden via AUWSX_DATA_DIR and AUWSX_SOCK environment variables.

auwsx is suitable for the following scenarios:

1. Long-running agent workflows (24/7 continuous codebase maintenance);
2. Multi-project parallel management (simultaneously monitoring and scheduling automated tasks for multiple codebases);
3. Human-machine collaborative development (agents handle routine tasks, humans focus on key decisions);
4. Auditable automation processes (complete state history and log records).

auwsx represents the transformation of autonomous coding agent infrastructure from single interaction to continuous operation. Through its daemon architecture, deterministic state pipeline, and clear human-machine collaboration interface, it provides a scalable, observable, and controllable autonomous work platform. Its application potential in scenarios such as automated code maintenance, intelligent code review, and continuous integration enhancement is worth looking forward to in the future.