This is an early release preview. You may encounter bugs.
Agent Manager logo
Unclaimed

Tool coding interface

Agent Manager

Terminal UI to manage AI coding-agent sessions (Claude Code, OpenCode, Codex, Grok, Gemini CLI) in tmux: live status, group tree, quick prom

A 80/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
No votes yet

01 / About

What Agent Manager is.

Agent Manager is a Go terminal UI over tmux for running several coding-agent sessions side by side. Every session appears in one list with live status, grouped into a project tree you can fold and reorder, so you see which agent is done and which is stuck without hunting through terminal tabs. It is a thin layer over the CLIs you already have: each session launches your installed tool as-is, with your login, subscription, config files, and MCP servers intact.

Most actions are a single key from the list. space sends a prompt into the selected session's pane as a user message, or spawns a new agent in the selected group; enter focuses a session in place; ctrl+r opens a full-file, syntax-highlighted diff of what an agent changed, where line comments go back to the agent's pane as one numbered review round. Sessions can be killed to free memory and revived on their own conversation, restarted on an empty context, or forked into a separately named conversation.

Sessions run on a private tmux server named agentmgr, so they survive the manager quitting and never mix with the tmux you run yourself, and you can reach one from a plain shell by naming that server. A session can spawn into its own git worktree on an am/ branch, from the new-session form, from the quick prompt, or by default. MCP tools let an agent work the same list: spawn a session, read another one's screen, message it, and wait on it before continuing.

Status detection ships for Claude Code, OpenCode, Codex, Grok Build, Gemini CLI, Pi, Command Code, and Hermes Agent, and any other CLI can run as a session or get its own status rules through a [tools. ] configuration block. It runs on macOS and Linux, and on Windows inside WSL2. Cost tracking and mouse-driven list navigation are listed as not present yet.

Features

  • One list, live status: every session's state in a foldable, reorderable project tree
  • Answer without attaching: a quick prompt bar sends replies straight into a blocked agent's pane, or starts a new agent in a group
  • Diff review: full-file diffs with line comments that return to the agent as a single review prompt, with sent comments marked open or handled
  • Worktree spawning: a session can take its own git worktree and branch, per launch or by default
  • Agents managing agents: MCP tools for spawning, reading, messaging, and waiting on other sessions, with file-overlap declarations on a shared checkout
  • Session lifecycle: kill, revive on the same conversation, restart on an empty context, or fork into a named conversation
  • Private tmux server: sessions persist across manager restarts and stay separate from your own tmux
  • Configurable tools: built-in status detection for eight CLIs, plus status rules and revive commands for any other tool

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

03 / Build

Build with Agent Manager.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Agent Manager →

Browse the catalogue