Convozen CLI: Build, Test and Deploy AI Agents from Your Terminal

To build a conversational AI agent, one generally has to deal with four separate interfaces: a prompt editor, a screen for configuring actions, a place where the knowledge base can be uploaded, and a settings panel covering voices, channels and variables. Each time you switch between these interfaces, you lose context. When a developer is testing a change to the prompt, they have to leave the editor, then reconfigure the action by pasting a curl command into a form, guess the shape of the parameters, and have to test the result without any guidance in a user interface that was designed for clicking rather than for building.

It isn’t a new issue. Over the past two years, enterprise engineering teams have incorporated AI-assisted workflows into their current tools rather than modifying their tools to be suitable for AI-assisted work. As Gartner points out, engineering organisations that have adopted AI coding agents have achieved a net average increase in productivity of about 19%. The same study observes that vendors are now competing less on the quality of code generation and more on their ability to coordinate complex workflows and integrate across engineering environments. The limitation has shifted from ‘can the AI write code’ to ‘can the AI work within the systems that a developer already uses’.

Convozen CLI was developed with that kind of change in mind; It is a terminal-first surface for creating, inspecting, editing, testing and publishing conversational AI agents, covering both the different types of AI agents a contact center might deploy, designed to run inside coding agent harnesses including Claude Code, Cursor, Codex and Antigravity, in addition to any standard terminal.

What the CLI Replaces

The Skills and the CLI collapse the process of building an agent into a single terminal conversation. Rather than switching back and forth between a prompt editor, an actions dashboard, a knowledge base uploader and a settings screen, the developer works from one location:

Getting Started: Install, Authenticate, Build

Setup is a two-step install followed by authentication:

1. Install the CLI since this will place the convozen binary on the PATH. It is available on macOS, Linux, and Windows (using Command Prompt or PowerShell, with automatic selection between x64 and ARM64).
2. Install the agent skills. You can add the Convozen skills to your preferred coding agent by running the command npx @convozen/convozen-skills install.
3. Log in to authenticate the CLI with your Convozen account.
4. Begin building right away via the CLI, or through Claude Code, Cursor, Codex or Antigravity after the skills have been installed.

For each supported agent, skills are installed in the standard skills directory, such as ~/.claude/skills, ~/.codex/skills, ~/.cursor/skills, ~/.antigravity/skills, ~/.gemini/skills, or a general ~/.agents/skills path for any agent that complies with the Agent Skills specification. Because agent configuration happens inline rather than in a separate console, response times during testing stay close to what the deployed agent will experience in production, which matters given how much latency shapes real-time AI conversations.

Key Capabilities of CLI

• Integration of native skills. Convozen provides ships with pre-configured skills that coding agents such as Claude Code, Cursor, and other supported coding agents can use directly, enabling the coding agent to carry out agent configuration without having to leave the terminal.
• Live backend connectivity. It is reported that the CLI integrates with a Model Context Protocol (MCP) server architecture so that the agents can access backend systems in real time, for example by retrieving order data, updating CRM fields, or connecting API endpoints.

Building WhatsApp Agents from the Terminal

The command-line interface is not only suitable for voice. It includes all of the WhatsApp Business features, so an agent created during one terminal session can operate across different modes:

• Text chat with multilingual, context-aware conversation across sessions
• Voice calls, inbound and outbound, with duplex speech.
• Voice notes, received, transcribed, understood and answered.
• Images, for catalogues, property photos or ID verification prompts
• Videos, for walkthroughs, explainers and onboarding
• Documents, for invoices and policies, including parsing of customer uploads

Template Studio for WhatsApp Messaging

In addition to agent logic, Convozen offers a studio that contains pre-approved templates for WhatsApp Business messages, including those used for marketing broadcasts, transactional updates and one-time passwords. The studio features five types of templates divided among three message categories: rich media headers (such as images, videos, documents or pinned locations), dynamic personalisation with up to ten variables in each message, interactive buttons allowing for quick replies and one-tap actions, and swipeable carousels with up to ten cards for use in catalogues and multi-offer campaigns. A live preview demonstrates precisely how the template will appear when rendered with sample data before it is sent out, and pre-approved templates are able to reach customers even outside the normal 24-hour messaging window.

The speed at which businesses now demand agent development makes this issue important. A CLI-based build process, where the agent, its actions and its customer-facing templates are all configured in one terminal session, removes the handoffs between writing agent logic and the tools needed to deploy an agent into a live contact center.

Who Convozen CLI Is Built For

Convozen CLI was designed for engineering and product teams who are already working within a coding agent or a terminal-based workflow and would like to incorporate agent configuration into that same environment, as opposed to keeping it separate in a web console; it is appropriate whether they are a single developer prototyping a WhatsApp flow or a platform team aiming to standardise how voice and messaging agents are built, tested and shipped throughout the organisation.