Overtime CLI
Agents that keep working without you: brief one once, and Overtime wakes it on a schedule and asks only what is yours to decide
- 402 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent runs as your user account and can write anywhere until you put paths in `protect`
- A background process keeps agents running after the app closes and spends up to 100 dollars a day per agent
- The agent asks before destroying things outside its folders, spending money or acting publicly, but the confirmation stays with you
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/overtimeDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent runs as your user account and can write anywhere until you put paths in `protect`
- A background process keeps agents running after the app closes and spends up to 100 dollars a day per agent
- The agent asks before destroying things outside its folders, spending money or acting publicly, but the confirmation stays with you
This is third-party code. Review the repository files before installing.
What it does
Overtime is a terminal app and a background process that keep agents working without you. You describe a job once in plain words, the agent writes it down in `AGENT.md`, plans the work, keeps notes and decides when to wake next: after a delay, on a schedule or on a signal from a watcher script. It asks only about what is yours to decide, with answer options and its own recommendation, and carries on with the rest. Overtime has no model of its own. It drives the coding agents you already use: the Claude backend is built in and works through your Claude Code login, Codex goes through an ACP adapter, and Gemini CLI and OpenCode connect when their CLI is installed and signed in. Any CLI that speaks ACP works too. Each Overtime agent lives in `~/overtime/agents/<name>/` with an `AGENT.md`, memory and its own skills. It splits big work between helpers, up to six at once, each in its own git worktree. A daily budget per agent limits spending, 100 dollars by default. Paths in the `protect` list become read-only for the agent and everything it starts, enforced by Seatbelt on macOS and bubblewrap on Linux. The `overtime acp` command connects Overtime to Zed, JetBrains, VS Code and Neovim as an ACP agent.
Who it is for. Developers and small teams who need standing agents to watch CI, handle incidents, review pull requests and prepare releases without starting sessions by hand.
Good fit when
- Dependency updates and CI break overnight, and a repo-keeper agent fixes the breakage and asks you before merging
- Metrics or logs spike with errors, a watcher script wakes the agent, and it finds the cause and rolls back the release
- Every pull request needs checking as it opens, with big ones split between helpers
- You want a daily review of papers, docs or competitors, and the report arrives only when there is news
- The work runs for several days: the agent keeps its own notes and wakes on a schedule
Not a fit when
- You need one edit per session: Claude Code or Codex alone is enough
- The agent must run on an API key without signing in to a coding agent: Overtime uses your Claude Code or other backend login
- You need full control of every step: the agent works on its own and asks only about what is yours to decide
Example request
You are repo-keeper: watch CI and dependency pull requests, fix what breaks and ask me before merging anythingLimitations
The project is early. The Claude backend is the best tested, and Codex and Gemini await wider testing. Node 22.19 or newer and at least one signed-in coding agent are required.
An agent can do whatever your user account can until you put paths in `protect`, so list protected folders beforehand. Overtime answers backend permission requests itself and instantly, so a session never hangs on a prompt.
Answering from a phone, a morning standup, shared agents and running on a server are listed in the README as future plans.
How to disable. Closing the app does not stop agents, since a background process keeps them running. The package comes off with `npm uninstall -g @ukanwat/overtime`, and agent data lives in `~/overtime/` and is deleted separately.
FAQ
How do I install and launch Overtime?
Run `npm install -g @ukanwat/overtime`, then `overtime`. Name the first agent, check the backend, model, daily budget and folder, press Enter and tell the agent what it is for. Node 22.19 or newer and one signed-in coding agent are required.
Which models does Overtime run on?
It has no model of its own: it drives backends. Claude, Codex, Gemini CLI and OpenCode are built in, and any ACP CLI goes under `customBackends` in `~/overtime/settings.json`. The model list comes from the backend, so new models appear without an Overtime update.
How do I limit spending and access?
The daily budget per agent is 100 dollars by default, changed in the agent settings or with `overtime set`. Folders the agent must not change go in `protect`, and they become read-only at the operating system level.
How does an agent ask me questions?
A question arrives with options and the agent's recommendation, pins above the message box and waits for your answer while the agent continues with the rest of its work. Each question and report also arrives as a notification. Only the thing asked about waits.
Can I connect MCP servers?
Yes. Servers under `mcpServers` in `~/overtime/settings.json` go to every agent, and a single agent gets its own in its `.overtime/settings.json`. Agents do not inherit your personal Claude Code settings or servers.
Discussion
No questions or reviews yet.
Related
CLI toolsA Rust CLI where a meta-agent hands work to Claude Code and Codex, each on its own branch and tmux session, in one view.
One runtime for Claude Code, Codex, OpenCode and more: a run counts as done only when checks the agent did not write pass
A multi-agent harness that verifies results by artifacts: stop hooks, an independent judge and an event log for Claude Code, Codex, Cursor and more
A memory and coordination layer for coding agents: specs and issues live in git while agents work in parallel
Local ledger of AI agent work: tokens, estimated cost, steps, changed files and checks per task for Claude Code, Codex and OpenCode
A tmux-based TUI that runs Claude Code, Codex and other agents in one list: status, quick prompts, worktrees and diff review
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| Overtime (this card) | no | 402 | |
| Ralph | no | 22,217 | loops agent runs with a clean context until the PRD stories pass |
| Agent Deck | no | 1,252 | manages sessions of several agents, and the agents do not wake on their own |
README in short
The README opens with an animation of a team of seven agents and states the idea: agents that exist, not sessions. A three-step quick start follows, ten example uses for work and for life, a brief of how it works (the agent folder, its own clock, questions, limits) and the project status. Details live in docs/how-it-works.md.