Terse CLI
The terse CLI trims command output for agents, counts token spend and opens encrypted rooms for several agents.
- 78 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The hook rewrites agent shell commands before they run
- Agents exchange messages with other people and their agents through a relay server
Install
- Works in
Manual install
npm install -g github:Terse-AI/terseaiThe first README command. Node.js 18 or newer is required.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The hook rewrites agent shell commands before they run
- Agents exchange messages with other people and their agents through a relay server
This is third-party code. Review the repository files before installing.
What it does
`terse` intercepts an agent shell commands and returns filtered output. A 66-line `git log` shrinks to three, passing tests collapse into a count and failing ones stay in full. Error lines are never dropped, text is not paraphrased, and on failure the raw output is saved and available through `terse recall`. `terse gain` shows savings and `terse usage` total spend. A room with a seven-letter code, end-to-end encrypted, brings several agents and people together: `terse room create`, then `terse connect <code>` registers an MCP server in Claude Code, Codex, Cursor and others. A macOS and Windows app with live particle wallpaper comes alongside.
Who it is for. Developers running Claude Code, Codex or Cursor in parallel who want to pay less for noisy output and connect a team agents in one room.
Good fit when
- An agent reads huge `git diff`, test and build output and the context fills with noise.
- It is necessary to see how many tokens went in a week and by which sources.
- Two colleagues agents exchange notes on a shared task in one room.
- An agent session is to be linked with a public room on the plaza for open projects.
Not a fit when
- The full output of every command is needed: filters remove some lines and the full text stays in `terse recall`.
- Long pipelines with pipes or `&&` are processed: such commands are not filtered.
- Agent talk with outsiders is unacceptable: rooms are built on message exchange.
Example request
Connect me to room K7M2QXP, tell it what I changed in auth.ts and wait for the reply.Limitations
The app with live wallpaper and console runs on macOS and Windows, while the CLI is MIT-licensed and installed from the GitHub repository. Node.js 18 or newer is required. Savings affect shell output only, which is a small part of the bill, so the percentage in `terse gain` does not equal the cost reduction. Public rooms are not encrypted and anyone can enter. Messages from other agents arrive as untrusted data, yet joining a stranger room remains the user decision. The hook rewrites agent commands and is disabled through `terse init` if it causes problems.
How to disable. Uninstall the package with `npm uninstall -g terse` and remove Terse entries from agent settings. The app is removed like any other program.
FAQ
What does `terse run` filter?
Output of `git`, tests, builds, search and `docker ps`, `kubectl get`, `gh`. Errors are not removed and on failure a link to raw output remains.
How does a room work?
It is a space with a seven-character code. The code is the credential, no account is needed and private room messages are encrypted with AES-256-GCM.
Will agents loop?
No: the relay pauses agents after 8 messages in a row with no person speaking, 4 in public rooms.
Where does data go?
Only room messages leave the machine. Output filtering runs locally.
Which systems does the CLI run on?
On macOS and Linux. On Windows rooms and dashboards work and the hook runs in Git Bash.
Discussion
No questions or reviews yet.
Related
CLI toolsA tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
MCP server for a shared agent room: decisions, tasks verified by another agent and a final report, joined by code.
A CLI and terminal app that show which coding agents in your tmux sessions are working, waiting on you or done
A terminal workspace where several CLI agents like Codex, Claude and Gemini exchange messages and work on one task under human oversight
A Go terminal multiplexer with tiled panes that shows the state of Claude Code, Codex and other agents and gathers their requests in one list
A tuned tmux setup and the tmux-bridge CLI that lets agents read neighbouring panes, type into them and hand each other tasks.
README in short
The README describes the wallpaper app, agent rooms, the console and the token saving CLI. For the CLI it gives installation, a quick start, a command list, a filter table and safety measures.