跳到正文
DEV Community· agent-manager·· 3 小時前AI 評分59

agent-manager 簡化 12 款編碼 Agent CLI 的上手流程,並匯入舊對話

Super easy onboarding for 12 coding agent CLIs, with your past chats imported

AI 導讀

agent-manager 0.45.0 為 12 款編碼 Agent CLI 提供引導式初次設定,並載入各 CLI 已儲存的對話。0.46.0 將訊息面板改為收件匣,並讓 Review 可逐個提交檢視分支。12 款工具各自在持續運作的 tmux 工作階段中執行,並在同一清單顯示即時狀態。

正文

agent manager is a local terminal app that connects to your existing subscriptions and tools. It runs Claude Code, Codex, OpenCode, Grok Build, Gemini CLI, Antigravity CLI, Pi, Command Code, Hermes Agent, Muse Code, Oh My Pi and Cursor CLI side by side, each in its own persistent tmux session, in one list with live status.

Version 0.45.0 makes the first five minutes the easiest part. One install command, then a short guided setup, and the list opens on the conversations your CLIs already saved instead of an empty screen.

Version 0.46.0, out now, turns the messages panel into an inbox and lets review step through a branch commit by commit.

Three agents spawned from the quick prompt bar, one each on Claude Code, Codex and Gemini CLI

Install

On macOS and Linux, install from homebrew-core. It brings tmux along if it is missing.

brew install agent-manager

For the prebuilt binary instead, use brew install --cask yoanwai/tap/agent-manager.

Without Homebrew, download the install script, read it, then run it. It verifies the release against the published checksums and installs to ~/.local/bin.

curl -fsSL -o install.sh https://raw.githubusercontent.com/YoanWai/agent-manager/main/install.sh &&
sh install.sh

On Windows, agent-manager runs natively with psmux. Scoop installs psmux with it.

scoop bucket add yoanwai https://github.com/YoanWai/scoop-bucket
scoop install yoanwai/agent-manager

An install.ps1 script and a verified zip are also available, and inside WSL2 it runs the same as on Linux. Arch Linux, mise and go install are covered in the install guide.

Then start it.

agent-manager

The first-run setup

A fresh install opens a short setup.

  1. Pick the agents you use.
  2. Pick a layout, list density and theme.
  3. Take a quick tour of the keys.
  4. Bring in the conversations your CLIs saved before.

Already installed? Press s for Settings and run the same setup from there.

Your tools, exactly as they are

agent-manager is a thin layer over the CLIs you already have. Each session launches your own installed tool as is. Your login, your subscription, your config files, your MCP servers and every feature the tool ships carry over, exactly as they behave in a plain terminal. The launch, revive, fork and status rules for all twelve tools ship in the binary, so one update brings the current version of each.

Sessions live on a private tmux server named agentmgr. They never mix with the tmux you run yourself, and they keep running when the manager quits.

A keyboard flow built for speed

Every session shows up in one list with live status, grouped into a project tree you can fold and reorder. No more hunting through terminal tabs to find which agent is done and which is stuck.

  • space opens quick prompt mode. Answer the selected session without attaching, or spawn a new agent in the selected group.
  • ctrl+t in quick prompt mode spawns the agent into its own git worktree.
  • ctrl+l, ctrl+x and ctrl+y pick the model, reasoning effort and profile, with the lists read from each CLI at runtime.
  • n opens the full new session form. enter focuses a session in place, and → and ← step in and out.
  • f continues a conversation in a separate named fork. R restarts a session on an empty context.
  • T opens a shell under an agent for builds, Git and one-off commands.
  • ? shows the key map for the current screen.

The mouse works across the whole manager as well. Click to focus, drag a row by its handle to reorder it, and open any row's actions from its menu.

Review mode

ctrl+r opens a full-file diff of what an agent changed, syntax highlighted. Press c to comment on a line, then C to send the comments back to the agent's pane as one numbered review round. Sent comments stay visible as open or handled, so feedback does not get lost between rounds.

Light on resources

Idle agents do not have to hold memory. x kills a session to free its RAM and keeps its row in the list. v revives it on its own conversation. A long list of agents can stay in view while only the ones you need right now are running.

Also new in 0.45.0

  • Environment profiles. Named sets of environment variables, such as a second CLAUDE_CONFIG_DIR, defined with agent-manager env and picked per session or as a group's default. One CLI can run on two accounts.
  • Config you can take along. agent-manager export and import move settings, presets, groups and SSH connections to another machine.
  • Resume on spawn. spawn --resume <id> and create_session with resume open a conversation a CLI saved earlier.
  • Reliable delivery. Prompts and messages count as sent only once the CLI takes them.

Agents that work together

Every MCP-capable session carries agent-manager's tools on launch, so an agent can spawn another, send it a message and wait until it is done. Saved prompts keep named instructions ready for the new session form, quick prompt mode, spawn --saved and send --saved.

Get started

agent-manager is free and open source under Apache-2.0. It runs on macOS and Linux, natively on Windows with psmux, and inside WSL2.

Using a CLI that is not on the list yet? Open a feature request on GitHub.

來源:DEV Community · dev.to