AgentX Overview

AgentX is an AI Assistant Management Platform for Kubernetes. It provides a web UI and REST API for creating, running, sharing, and administering AI coding assistants — each deployed as an isolated Kubernetes pod with a persistent workspace.

What is an Assistant?

An assistant is a Kubernetes StatefulSet pod that contains:

  • A persistent workspace (PVC) pre-populated from a git repo or blank

  • Seven in-browser coding-agent terminals: Claude Code, OpenCode, Gemini CLI, Codex CLI, GitHub Copilot CLI, Mistral Vibe, and a plain shell

  • An OpenClaw gateway serving the assistant’s web UI

  • All processes managed by Supervisor, started lazily on first browser interaction

Users access their assistant through a unique URL (/agentx/<uuid>/) and interact with agents entirely in the browser — no local CLI setup required.

Key Features

Assistant Management

  • Create & deploy — launch assistants from scratch or from a template in one click

  • Workspace sources — optionally clone a git repo (SSH) into the workspace at creation time, or start blank

  • Lifecycle control — start, stop, restart, and delete assistants; status reflected in real time

  • Real-time logs — stream pod logs directly in the UI via Server-Sent Events

Coding-Agent TUIs

Each assistant pod exposes seven browser-accessible terminals:

ID

Label

Description

shell

Terminal

Plain bash shell in the workspace

claude

Claude Code

Anthropic Claude Code CLI

opencode

OpenCode

anomalyco/opencode CLI

gemini

Gemini CLI

Google Gemini CLI

codex

Codex CLI

OpenAI Codex CLI

copilot

Copilot CLI

GitHub Copilot CLI

vibe

Mistral Vibe

Mistral Vibe coding agent

Terminals are lazily activated — a process starts only when the user first opens that terminal tab.

Collaboration

  • Share assistants with other users; three permission levels: VIEW, USE, MANAGE

  • Pin favorites for quick access across sessions

  • Shared with Me tab shows all assistants shared with you

Templates

  • Publish any assistant’s configuration as a reusable template

  • Version control — create and deprecate template versions

  • One-click deploy — create a new assistant from a template

Administration

  • User management — create, update, and delete users; assign admin roles

  • System monitoring — health overview, pod CPU/memory metrics

  • Audit log — record of all admin actions

How It Works

1. Create Assistant  →  2. Workspace Set Up  →  3. Use in Browser  →  4. Share & Collaborate

  1. Create — submit a name, description, provider/model, and optional workspace source (blank or git repo)

  2. Deploy — AgentX renders Kubernetes manifests and applies them; the pod reaches running status once the readiness probe passes

  3. Use — open the assistant URL; agent terminals lazy-start on first interaction; workspace persists across pod restarts

  4. Share — share with teammates at VIEW / USE / MANAGE permission; they see it in “Shared with Me”

Technology Stack

  • Frontend: React 18 + TypeScript, TanStack Query, Tailwind CSS, Vite

  • Backend: FastAPI (Python 3.12+), SQLModel, PostgreSQL (production) / SQLite (development)

  • Assistant runtime: Node.js 24, Jupyter terminal servers (terminado), OpenClaw gateway, Supervisor

  • Orchestration: Kubernetes StatefulSets, Gateway API HTTPRoutes, Helm

  • Authentication: OAuth2 Proxy (Kubernetes) / local-admin (development)

  • Model provider: DKubeX (SecureLLM gateway)

Use Cases

Development Teams

  • Shared, persistent coding environments for every project

  • Standardize tooling with templates

  • Pair-program by sharing an assistant

AI/ML Projects

  • Deploy multiple assistants for different experiments

  • Version and share configurations via templates

  • Use multiple coding agents side-by-side in the same workspace

Enterprise

  • Centralized assistant lifecycle management

  • Role-based access control and audit trail

  • Resource usage monitoring per user