Hive CLI
A local browser workspace for Claude Code, Codex and other CLI agents where one agent hands out tasks to the rest
- 579 ★ on the repository
- Last commit
Broad access
Why this level
- Workers run in a mode without approval prompts and execute shell commands with your account's rights
- There is no sandbox or multi-user authentication, and every agent gets access to the whole working directory
- The optional remote access from a phone exposes the full Hive interface with the same rights as the local browser
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/hiveDetects 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.
- Workers run in a mode without approval prompts and execute shell commands with your account's rights
- There is no sandbox or multi-user authentication, and every agent gets access to the whole working directory
- The optional remote access from a phone exposes the full Hive interface with the same rights as the local browser
This is third-party code. Review the repository files before installing.
What it does
Hive launches several CLI agents on your computer as ordinary processes with a real terminal and shows them in one browser window. One agent becomes the orchestrator: it plans the work and hands tasks to workers. The workers write code, review changes and test, then report back to the orchestrator. Inside the agent sessions Hive adds a `team` command. The orchestrator assigns a task with `team send`, and a worker hands in the result with `team report`. The shared task list sits in a `.hive/tasks.md` file in the working directory, and you can read and edit it outside the app. Ready presets cover ten CLI agents, including Claude Code, Codex and Qwen Code. The Workflows mode is off by default: it lets the orchestrator run multi-step scenarios with several agents.
Who it is for. For developers who already work with CLI agents and are tired of switching between their terminals.
Good fit when
- You want one agent to implement a fix and another to check edge cases before the final report
- You want to split a hunt for an intermittent bug between the server side, the interface and recent commit history
- You need a technical note: one agent gathers facts, a second writes, a third checks commands and file references
- You want to see in one window what every agent is doing
Not a fit when
- One agent in one terminal handles the task: the extra layer only adds processes
- You need a sandbox or separated users: Hive does not isolate agents and has no multi-user authentication
- The repository holds secrets or code you do not trust: workers get access to the whole working directory
- The system runs Alpine or another musl build, or an architecture other than x64 and arm64: there are no prebuilt binaries for it
Example request
Ship the settings search bugfix. Use one worker to implement it and another to review edge cases before the final report.Limitations
You need Node.js 22.18 or newer on the 22 line, or version 24 and up. You also need at least one installed and signed-in CLI agent. macOS and Linux get first-tier support, and Windows comes second. CI runs a minimal compatibility check for Claude Code and Codex, and the author tests the other presets by hand. So changes in the agents themselves sometimes reach Hive with a delay. The license is BUSL-1.1: the source is open, but the project does not meet the OSI definition of open source. Personal and internal use is allowed. Offering Hive to third parties in a competing product is not. Each version moves to Apache 2.0 on May 16, 2030 or four years after its first release, whichever comes first.
How to disable. Stop the `hive` process and remove the package with `npm uninstall -g @tt-a1i/hive`. Metadata lives in `~/.config/hive` (`%APPDATA%\hive` on Windows) or in the folder from the `HIVE_DATA_DIR` variable, and task lists in `.hive/tasks.md` of your workspaces: delete them separately. If you turned on remote access, revoke the paired devices.
FAQ
How do I install and start Hive?
Run `npm install -g @tt-a1i/hive` and then `hive`. Open the local address from the output, usually http://127.0.0.1:9483/. The `--port 0` flag asks the OS for a free port, and `hive update` upgrades the install in place.
Do I need to install Claude Code or Codex separately?
Yes. Hive does not install CLI agents. The command must be on PATH in the same shell you start `hive` from, and the agent must be signed in.
Can I try Hive without any agents installed?
Yes, the first-run wizard has a Try Demo button. The demo runs entirely in the browser: a simulated terminal goes through planning, `team spawn`, `team send` and reports. No network, installed agent or sign-in is needed.
What is risky about the mode without approval prompts?
The built-in presets launch agents in bypass mode where it exists: `--dangerously-skip-permissions` for Claude Code and `--dangerously-bypass-approvals-and-sandbox` for Codex. A worker can run any shell command in the chosen workspace with your account's rights.
Can I use Hive from a phone?
Yes, if you turn on the optional remote access, which is off by default. You need a gateway that you stand up yourself, for example on Cloudflare Workers. You sign in through GitHub or Google with `hive remote login`, and each new device is confirmed at the computer.
Discussion
No questions or reviews yet.
Related
CLI toolsOpen-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
An engineering system for coding agents: 68 agents, 292 skills, hooks, memory, rules and the AgentShield security scanner
DeepSeek Harness is an open-source AI agent from DeepSeek where every feature is a plugin. It runs in the browser, on macOS and on Windows.
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Ollama runs open models on your own machine and serves them to agents through a local API compatible with OpenAI and Anthropic.
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| Hive (this entry) | no | 579 | |
| Claude Squad | no | 8,656 | a terminal app for parallel sessions in tmux and git worktrees, with no shared task list or orchestrator |
| oh-my-claudecode | no | 40,191 | an orchestration plugin inside Claude Code; Hive runs different CLI agents as separate processes in a browser |
| amux | no | 544 | a control plane that runs Claude Code |
README in short
The author starts with why to coordinate several CLI agents and walks through three scenarios: a fix with review, a parallel bug hunt and work on a text. Separate sections cover the safety model and where Hive keeps its data.