A lightweight terminal chat agent built on the OpenAI Responses API through the official @eliware/openai client.
- Features
- Requirements
- Setup
- Usage
- Development
- Testing
- Troubleshooting
- Security
- Configuration
- Operations
- Commands
- Exit codes
- Support
- Links
- License
- Send messages to the OpenAI Responses API from a terminal.
- Run local shell commands with the
!prefix. - Use optional MCP tools and worker agents.
- Resume saved sessions and restore successful checkpoints.
- Node.js 26 or later.
- An OpenAI API key.
- Windows or Linux.
Install AgentX and its setup command:
npm install --global @eliware/agentx-cli@latest
agentx-setupStart an interactive session:
agentxSend one request and exit:
agentx "summarize this project"Read the release notes for version changes. Read the user guides for supported workflows.
Read specs/ before you change product behavior. Use Node.js 26 and native ESM.
Run all required checks with:
npm testRun agentx-setup if AgentX reports a missing API key. Run agentx --help to view options. Read the troubleshooting guide for other issues.
Permission labels do not create a security boundary. Use --confirm to review high-risk tool calls. Keep API keys and user data private.
AgentX reads runtime settings from user files and supported environment variables. Use /setup to edit runtime settings. Read the configuration guide for details.
AgentX checks settings before network access. It contacts OpenAI after your first message. Use quit, exit, /quit, or /exit to close a session.
Use cd PATH to change the working directory. Prefix a local shell command with !. Use /usage, /rollback, /setup, or /clear in a session. Read the command guide for all commands and options.
Code 0 means success. A nonzero code means failure. Errors do not show credentials.
For help, questions, or to chat with the author and community, visit:
