AgEnD CLI
A local daemon assembles Claude Code, Codex and OpenCode into a team: tasks, separate worktrees, checks and merging.
- 64 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Several agents write code, run checks and merge changes in Git without constant supervision.
- Permission and merge approvals can be given from Telegram, and the bot token opens control of the agents.
Install
- Works in
- Services
Manual install
git clone --branch main https://github.com/suzuke/AgEnD.git agend-v2The README's first step. Then `cargo install --locked --path crates/agend --root "$HOME/.local"`. Git 2.38, Rust through rustup and compilers are required.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Several agents write code, run checks and merge changes in Git without constant supervision.
- Permission and merge approvals can be given from Telegram, and the bot token opens control of the agents.
This is third-party code. Review the repository files before installing.
What it does
AgEnD runs an `agend` daemon on the machine that ties the work of several agents into one flow: task creation, development, checks, review, merge. Claude, Codex and OpenCode agents can be gathered into one team, given roles and asked to review each other. Each task runs in its own Git worktree, and the daemon coordinates merging. In the terminal interface the "Needs you" section gathers permission requests, questions and reviews in one place, and decisions can also be made from Telegram. The interface opens any agent's native terminal, and you can take over input. A separate holder process keeps agent processes, so they are picked up again after the daemon restarts.
Who it is for. Developers who run several agents in parallel on one project and are tired of switching windows for approvals and reviews.
Good fit when
- Claude and Codex must split tasks by role and review each other's work.
- Several agents write code at once and each needs its own Git worktree without conflicts.
- You want agent permission requests and questions in one list rather than a dozen windows.
- Merge approval is tied to a commit and re-evaluated after any change.
- An agent's action must be approved from a phone through Telegram.
Not a fit when
- You work with one agent in one window: the daemon and interface add surplus.
- You need a prebuilt binary: installation currently runs from source, with no releases or Homebrew.
- Windows: only macOS and Linux are supported.
Example request
Create a team of Claude and Codex, assign them a module refactoring task and show me the review once the checks pass.Limitations
The project is a trial version: installation runs only from source through `cargo install`, with no public releases or Homebrew. Git 2.38 or newer, Rust through rustup and macOS or Linux build tools are required, and the project selects Rust 1.96.0 itself. The interface and documentation are mainly in Traditional Chinese, and the language can be switched. Communication with Claude is connected, but acceptance of real upgrades and rollbacks of managed model versions is postponed. Agents work without constant supervision, so permissions and merge rules must be set up in advance.
How to disable. Stop the daemon in the terminal where it runs and delete the binary from `~/.local/bin`. Remove the data directory set by `AGEND_HOME` and the working worktrees by hand.
FAQ
How do I install AgEnD?
Clone with `git clone --branch main https://github.com/suzuke/AgEnD.git agend-v2`, then run `cargo install --locked --path crates/agend --root "$HOME/.local"` and add `~/.local/bin` to PATH. Verify with `agend --version`.
How do I start it the first time?
In the first terminal run `agend init` and `agend daemon`, and wait for `agend daemon ready`. In the second add an agent with `agend instance add try-opencode opencode` and open the interface with `agend app`. The `i` key takes over input.
Which agents does AgEnD work with?
Claude, Codex and OpenCode. They can be gathered into one team, given roles and set up for mutual review. The interface runs in a terminal on macOS and Linux, and notifications and approvals are mirrored in Telegram.
How do I view the interface without running models?
In the source directory run `cargo run --locked -p agend-tui --example tui_fake -- --lang zh-TW`. A demo interface opens on test data with no sign-in and no models started.
What if something goes wrong?
Run `agend doctor` and make sure both terminals share the same `AGEND_HOME` value. Other steps are in the quickstart document under troubleshooting.
Discussion
No questions or reviews yet.
Related
All Telegram toolsConnects Claude Code, Cursor, Codex or OpenCode to Telegram, Slack and Discord: teammates message a bot, an agent replies
Claude-to-IM
claude-to-im
A bridge between Claude Code or Codex and messengers: talk to the agent from Telegram, Discord, Feishu, QQ or WeChat
A compact OpenClaw alternative: a personal assistant on Claude Agent SDK with agents in isolated containers and messengers added via skills
A CLI to run several coding agents: one config, a session console, local memory and a set of engineering skills
Self-hosted monitoring and remote approvals for Claude Code and Codex via Telegram, Feishu or a web terminal
A Node.js terminal agent that takes a goal, works with files, shell and browser and backs its result with evidence
README in short
The authors put the idea this way: agents work as a team and the human decides only what needs deciding. The end-to-end scheme runs from task creation through checks and review to merge.