Agent Manager is an open-source terminal-based agent harness and session orchestrator developed by Yoan Wai. It is built to run, monitor, and manage the execution of multiple command-line AI coding assistants from a unified control interface, tracking agent lifecycles across terminal panes. In daily use, Agent Manager acts as a supervisor over CLI agents such as Command Code (`cmd`), `pi`, and other terminal assistants. It continuously monitors TUI outputs using configured status rules, inspecting pane elements such as composer prompts, spinners, turn-completion markers, and approval dialogues. This allows the harness to automatically track state transitions across waiting, working, tool approval, and finished states without interfering with the underlying agent. What distinguishes Agent Manager is its resilient terminal state-matching engine and session tracking. Rather than requiring agents to conform to proprietary telemetry APIs, it uses live terminal capture parsers to detect tool approval prompts, error states, and rate limit banners, alongside an integrated SQLite-backed session store that supports reviving and resuming exact conversation threads across agent restarts. Agent Manager is open-source software distributed via GitHub. It is implemented with lightweight local dependencies, including local SQLite storage and clipboard integration, making it straightforward to run in terminal-first developer environments.
Tags: agent harness, cli, coding agent, developer tools, terminal