Vigil
Vigil is an open-source native macOS multi-agent terminal orchestrator. It organizes AI coding agents into a manager-driven tree: a manager agent spawns workers and sub-managers, with detailed task context flowing down the hierarchy and summaries rolling back up. Each agent node runs inside its own terminal user interface (TUI), so users can inspect or take over any node directly. The project is designed for terminal-based AI coding CLIs such as Claude Code, Codex, and OpenCode. Rather than forcing all agents into a single chat or hidden process, Vigil keeps each agent in the CLI's native TUI, enabling interactive control and transparent monitoring. It is distributed as an open-source GitHub repository and targets macOS users who want to coordinate multiple coding agents. Because it is native to macOS and terminal-centric, Vigil is best suited to developers already using command-line AI agents. It provides a structured way to decompose larger coding tasks into subtasks, delegate them, and collect results without losing the ability to intervene manually.
- Category
- AI Agents
- Pricing
- Open Source
Problem solved
Managing multiple AI coding agents in terminals is manual and error-prone: users must open separate panes, track context, copy summaries, and intervene without a clear hierarchy. Vigil solves this by providing a manager-driven agent tree where tasks and context flow down, summaries flow up, and every agent remains an interactive TUI that can be taken over.
Use cases
- Coordinate multiple AI coding agents such as Claude Code, Codex, and OpenCode from one macOS terminal.
- Decompose a complex coding task into a tree of worker and sub-manager agents.
- Monitor and take over any running agent's TUI to provide manual corrections.
- Aggregate summaries from worker agents into higher-level manager reports.
- Run parallel coding experiments or refactors with hierarchical agent oversight.
- Prototype and study multi-agent orchestration workflows locally.
- Keep interactive control over AI-generated code changes while scaling agent delegation.
Audience
- Developers
- ML Engineers
- Researchers
- Hobbyists
- Startups
Pros
- Open source and locally runnable on macOS.
- Manager-driven agent tree enables hierarchical task delegation.
- Each agent stays in its native CLI TUI, allowing direct user takeover.
- Supports multiple popular coding agents such as Claude Code, Codex, and OpenCode.
- Provides structured context flow down and summary roll-up across agents.
Cons
- macOS-only; no native Windows or Linux support indicated.
- Requires familiarity with terminal-based AI coding CLIs.
- Depends on external agent CLIs and their authentication/setup.
- Early-stage project with limited documentation and community evidence.
- No built-in GUI; all interaction is terminal-centric.
Official website: https://github.com/butterlatte-zhang/vigil
Canonical LambdaBase page: https://www.lambdabase.com/tools/vigil