Zerg Client

Zerg Client, also referred to as ZTC or Zerg Terminal Client, is an AI coding agent for the terminal and the desktop. According to the product site, its core job is to plan a change, edit code, and run tests in your project. Zerg AI describes it as a coding agent harness that drives its own terminal panes: when you ask for Vim or htop, the agent opens them in panes beside it. Those panes can also hold more agents, each running on its own model. The client operates against a real project directory — the site's example is ~/acme-api on the main branch, marked clean — and shows runtime information such as "runtime: not configured" next to the version, currently ZTC v0.9.0-beta.1. Zerg Client is presented as an evaluation beta and ships both as a terminal CLI and as a desktop app. The context for Zerg Client comes from how it is packaged and what it targets. ZTC is aimed at developers who already work inside a project directory in a terminal, following a loop of planning a change, editing code, and running tests. Zerg AI frames the product as a harness rather than a single-purpose utility: the harness drives terminal panes, which means interactive programs a developer relies on — the site names Vim and htop — can be opened in panes beside the agent. That detail matters because those are full-screen terminal applications that normally live inside a shell session, so an agent that can open them keeps the developer's existing terminal tooling in view. The product is also flexible about which model powers the agent: available models depend on the sign-in route and provider access, and panes can hold more agents, each on its own model. Terminal pane orchestration is the defining capability of Zerg Client. Rather than confining the agent to a single text interface, the harness opens terminal panes beside itself and drives them. The site gives two concrete examples of what can be launched this way — Vim and htop — and the Product Hunt listing explains that you simply ask for Vim or htop and the agent opens them in panes beside it. Because these panes are ordinary terminal surfaces, they can also host additional agents, so a single session can grow into a small workspace of side-by-side agents. That structure lets a developer keep the agent's work, an editor, and a system monitor or other terminal tool visible at the same time instead of switching between separate windows. Connecting a model is handled by a dedicated setup step titled "Connect a model," which asks how ZTC should reach a model and offers two routes. The recommended route is logging in with ZergAI, which opens your browser to log in and provides a single ZergAI login for ZTC in both the terminal and the desktop app. With this route, provider keys are added once in Zerg Router rather than on the local machine, and DeepSeek is ready on Zerg Router billing with no DeepSeek key needed. Router billing also provides usage history, per-key daily budgets, and a monthly cap. The alternative route is to use your local API keys. Whichever route is chosen, models can be switched at any time with the /model command, and the client's status line reflects the current state — for example, showing "no model" until a connection is configured. Installing the terminal client is done through a package manager or a direct script. The site lists npm, bun, and curl options, with the npm command shown as npm install -g @zergai/ztc@0.9.0-beta.1. The CLI requires Node.js 20.19+ or 22.12+, and after installation you run ztc in your project directory to spin it up. A desktop app is offered as an alternative, currently available for macOS on Apple Silicon as version 0.9.0-beta.1; the Windows and Linux desktop apps are listed as coming soon and are explicitly noted as not available yet. The site also distinguishes native terminal support (macOS, Apple Silicon) from headless support (macOS, Linux, and Windows), and mentions an optional browser mirror provided as a separate ztc-web-client plugin. Overall, the workflow Zerg Client describes runs from authentication to model connection to project work. You install the terminal CLI or the desktop app, optionally run ztc auth login first to skip the prompt, then choose how to reach a model — either the ZergAI login or your own local API keys — and finally point the client at your project directory, where it can plan a change, edit code, and run tests. From there the harness takes over the terminal: it can open tools such as Vim or htop in panes beside it and place additional agents in further panes, each with its own model. The Quickstart and Benchmarks links on the site point to further documentation and performance results for users who want detail beyond the initial setup. Several benefits follow directly from that design. Because a single ZergAI login covers ZTC in the terminal and the desktop app, users do not need to maintain separate credentials for each surface. Because provider keys can be added once in Zerg Router rather than on the machine, key management is centralized, and DeepSeek works on Zerg Router billing without a separate DeepSeek key. Usage history, per-key daily budgets, and a monthly cap give visibility and spending limits on Router billing, while /model lets a user change models at any time without reconfiguring the client. And because the agent drives its own panes, the tools it launches sit beside it rather than in another window. Concrete usage scenarios come straight from the material. A developer working in a project such as ~/acme-api can ask the agent for Vim and get an editing pane beside the agent, or ask for htop and get a system monitor in a neighboring pane. The same pane structure supports running several agents at once, each on its own model, which suits work where you want different models applied side by side. The core loop — plan a change, edit code, run tests — describes day-to-day project work in a terminal. Teams that prefer centralized key handling can connect models through Zerg Router, while developers who already hold their own provider keys can use the local API keys route instead. The optional ztc-web-client plugin adds a browser mirror for those who want one. Zerg Client targets developers who work in terminals and project directories, and Zerg AI's own framing places it in developer tooling and software engineering. It runs as a terminal CLI installed through npm, bun, or curl with Node.js 20.19+ or 22.12+, and as a desktop app for macOS on Apple Silicon, with Windows and Linux desktop apps coming soon. Headless use is supported on macOS, Linux, and Windows. On cost, the site states that ZTC is free to evaluate under the Zerg Source Available License, that commercial use requires a license, and that model usage is billed separately; Zerg Router billing adds per-key daily budgets and a monthly cap. In short, Zerg Client's value proposition is an AI coding agent that works where the developer already works — the terminal and the desktop — and that plans changes, edits code, and runs tests while driving its own terminal panes. With a choice between a single ZergAI login backed by Zerg Router billing and your own local API keys, per-pane agents on separate models, and a CLI plus macOS desktop app in evaluation beta, it is built around keeping the agent and the developer's terminal tools side by side.